Platform first-party AI

Proven rule. Every body claim traces to a named source with an accessed date. Amazon's and noon's own domains outrank trade press, which outranks agency blogs. Inferred, single-sourced-shaky or contradicted-without-a-winner claims are in §4, not §1. Where a researcher was blocked, §4 says so — a tooling limit is never reported as absence.

---

Verdict

Amazon has shipped a great deal of free seller AI and almost none of the agentic layer has reached the Gulf. Noon has shipped one button.

As of 2026-09-06: Amazon's agentic Seller Assistant is documented on .com, .co.uk/EU, .in and (in its older chat-only form) .co.jp, and is not present in the amazon.ae or amazon.sa help catalogues against a working control on the same hosts [A2-S13]. Amazon's free AI ad-creative suite — Creative Agent, image, video and audio generators — publishes four explicit country lists and the UAE and Saudi Arabia appear in none of them, including on Amazon's own Arabic-UAE localisation of that page [A2-S32] [A2-S33]. What has landed in the Gulf is narrow and documented: AI listing enhancement ("View enhancements" / Enhance My Listing) is live and per-store documented on .ae and .sa [A2-S21] [A2-S22]; Ads Agent is flagged available for UAE and KSA but gated behind Amazon Marketing Cloud + Amazon DSP and an account executive [A2-S26] [A2-S27]; the Amazon Ads MCP Server is "available globally" to Ads API partners [A2-S34].

noon documents exactly one AI feature for sellers — a category classifier in Seller Lab's listing flow — across 243 English help-centre articles and the whole partner-API index [A2-S65] [A2-S66] [A2-S64] [A2-S74]. It also states in writing that it will not translate seller content [A2-S67].

Amazon's posture toward third-party AI is not hostile; it is contractual. Amazon publishes an LLM-addressed page naming Claude [A2-S44] and an MCP server naming Claude, ChatGPT and Gemini [A2-S34], charges no listing fee and takes no revenue share on Appstore apps [A2-S56] — and simultaneously binds any "software or service that takes autonomous or semi-autonomous action" under §1.3 Use of Agents of the UAE-locale developer agreement [A2-S56], caps an unlisted public app at 25 seller authorizations [A2-S54], and raised the vendor security floor on 2026-08-25 [A2-S60].

The escalation trigger in the Area Brief did not fire, and it has a 16-day shelf life. Amazon Accelerate 2026 runs 22–24 September 2026 with the author of the Seller Assistant press release as a featured speaker [A2-S12]. Accelerate is where Amazon announced this product class in 2023, 2024 and 2025. Anything in §1.2 and §2.1 must be re-checked after 2026-09-24.

---

Source legend

All URLs accessed 2026-09-06 unless a different date is stated. Type: P = the platform's own domain (press, help, docs, contract, API); T = trade press; X = vendor/agency blog, treated as claim not evidence; O = this Inquiry's own reproducible probe.

Amazon — Seller Assistant

Amazon — listing and A+ content AI

Amazon Ads

Amazon — partner posture, Appstore, developer contracts

noon

Cross-Inquiry references

006:[L09], 006:[L18], 006:[L20] — Inquiry 006 finding 07. A1:[A1-S38], A1:[A1-S35] — the competitor blog and tools index behind A1's unresolved Seller Assistant note. A6:[A6-S24], A6:[A6-S26], A6:[A6-S32], A6:[A6-S34] — Ads MCP access mechanics, the March 2026 BSA announcement, and the noon API surface.

---

1. What we found

1.1 Headline: the Gulf gets a thin slice of Amazon's AI, on a long lag

Amazon has been shipping seller-facing generative AI since Accelerate 2023 [A2-S18] and agentic seller AI since Accelerate 2025 [A2-S1]. Almost all of it is free. Very little of it is in the Gulf.

The pattern is consistent enough to be a rule, and it is Amazon's own documentation that establishes it, not inference:

• Listing-content AI reached .ae and .sa. The help article that documents the "View enhancements" AI panel carries an explicit per-store scope line and renders on all three storefronts: "This article applies to selling in: United States / United Arab Emirates / Saudi Arabia" [A2-S20] [A2-S21] [A2-S22], with a byte-identical AI section.
• Agentic assistance did not. The Seller Assistant help article is absent on .ae and .sa against a control article that renders in full on the same hosts the same day [A2-S13], and Amazon's own launch sentence is "currently available to all sellers in the U.S. store and will be rolled out to other countries in the coming months" [A2-S1] — written 2025-09-17, i.e. ~12 months ago.
• AI ad creative did not, and Amazon says so in Arabic. The Creative Solutions page footnotes four country lists; UAE and Saudi Arabia appear in none [A2-S32], and the Arabic-UAE localisation of that same page carries the same lists [A2-S33].
• The A+ gen-AI wave took ~17 months to leave the US (US ≈Sept 2024 [A2-S23] → Europe/Canada/India ≈Feb 2026 [A2-S24]) and the Gulf was in neither wave — though see §4 P5, this one is a failure to find, not a demonstrated absence.

The one place Amazon is unambiguously global is the developer surface: the Ads MCP Server is "available globally in open beta to Amazon Ads partners with active API credentials" [A2-S34] — the strongest availability language found anywhere in this Area, and it points at third parties, not at sellers.

1.2 Seller Assistant: what it is, that it writes, and where it does not run

The name is Seller Assistant. "Project Amelia" was the Accelerate-2024 codename [A2-S19]; Amazon's 2025 press release refers back to the 2024 launch as "Seller Assistant" with no mention of Amelia [A2-S1], and every current Amazon surface uses only "Seller Assistant" [A2-S2] [A2-S3]. (Naming trap: sellerassistant.app is an unrelated third-party arbitrage-sourcing extension [A2-S9].)

It is free. Amazon says so twice: "Is the Seller Assistant free or is there a fee to use it? It's free to use." [A2-S3], and "…will be rolled out to other countries in the coming months, at no additional cost" [A2-S1].

It is built on Amazon's own stack. "Powered by Amazon Bedrock, and leveraging Amazon Nova and Anthropic Claude" [A2-S1].

It writes to the account, gated by a Seller Central user permission plus per-action approval. Amazon documents a permission literally named "Seller Assistant" with the standard None / View / Edit levels, administrator-grantable, and warns: "Only grant access to trusted individuals, as this access allows secondary users to view account data and they can make account-level, ASIN-level, or other changes" [A2-S3]. One of the four documented prompt classes is a write — "Update my product title, description, and bullet points" [A2-S3]. The agentic action list, verbatim: "investigate inactive or blocked listings, help resolve stranded inventory, analyze storage fees and aged inventory surcharges, assess Featured Offer eligibility, identify open compliance violations, and take actions on your behalf when you authorize it to do so" [A2-S3].

The trajectory is explicitly toward more writing. The canvas experience, published 2026-03-02 and "built on Seller Assistant's agentic architecture", says: "Once sellers choose a path forward, they can work with Seller Assistant to take action—and over time, Seller Assistant will execute more of these actions on behalf of sellers, like updating prices or creating restock orders" [A2-S4].

And Amazon has drawn its own scope boundary, in a disclaimer that is the single most useful sentence in this Area for the wedge argument:

"Do not rely on this tool for professional, legal, or financial advice or attempt to use the assistant for answers to questions not related to selling in the Amazon store." [A2-S3]

Trade licence, sourcing, freight, customs and importer-of-record, VAT registration, ESMA/SFDA conformity, Arabic labelling — the entire pre-Amazon half of the SCOPE spine — is outside that boundary by Amazon's own definition. The one caveat: the announced compliance-navigation capability does cover product compliance documents (Amazon's example is a missing UL certification on an electronics listing) [A2-S1], i.e. the listing-facing slice of compliance, not the importer's regulatory slice.

Where it runs, 2026-09-06. Method and result in §2.1. Present and agentic on .com, sellercentral-europe (UK/EU) and .in; present but titled "conversational assistant" — the older, pre-agentic generation — on .co.jp; absent on .ae, .sa, .com.au and .sg against controls that render [A2-S13]. Egypt is inconclusive because the control also failed there.

That Japan gradient matters: availability is not binary. A store can receive the 2024 chatbot years before it receives the 2025 agent.

Adoption, as Amazon publishes it. "more than 230,000 monthly users" and "Independent sellers accept the recommended actions of the AI-powered Seller Assistant over 90% of the time" [A2-S2], the latter attributed on a second Amazon page to the 2025 Amazon Small Business Empowerment Report [A2-S44]. Neither figure carries a date on Amazon's page — cite as "as published, accessed 2026-09-06", never as a dated 2026 metric (§4 P8).

1.3 The one first-party AI capability that is genuinely live in the Gulf

Enhance My Listing / "View enhancements" is documented, per-store, on amazon.ae and amazon.sa. This is the strongest positive Gulf finding in the Area and it was proven, not assumed: help article G200335450 was rendered on all three storefronts on 2026-09-06, each carrying its own store-scope line, each containing the identical AI section [A2-S20] [A2-S21] [A2-S22]:

"When suggesting changes to your product listing, use View enhancements to review any available AI-generated suggestions that help keep your product content accurate, relevant, and up to date… Note: It's your responsibility to review and verify all generated content thoroughly… AI-generated content can include features and claims that weren't in the listing's original content."

Three readings follow.

First, it is free and it is inside the edit flow. A third party cannot beat "already in the button you were about to click, at no cost."

Second, it is suggest-and-review, not autonomous. Amazon ships the draft and hands the seller the verification risk, in writing. That is exactly the watch, advise, draft ceiling SCOPE accepts for v1 — Amazon's own first-party tool does not exceed it.

Third, sellers report it is not good. The US announcement thread is the only field evidence of quality found [A2-S15], and all three top-voted replies are negative — verbatim:

"Would be useful if the tool actually showed what changes it was proposing and comparing them to the existing content… there is no proof that the tool increases sales, such as A/B tests" (5 upvotes)

"the AI's only recommendation was to restore the sort of faulty title it had previously forced on us — that we had to fix manually" (5 upvotes)

"one of them (model name) showed 'mistral-3-1' which has nothing to do with what I sell. Isn't mistral the name of an LLM? Why is it being shown in my listing as an 'enhancement'?" (4 upvotes)

Three replies on one thread are anecdote, not a quality measurement. But they are Amazon-hosted, dated, and they define the shape of what is not foreclosed: a tool that evaluates Amazon's own suggestions before the seller publishes them is a different product from a tool that generates suggestions.

1.4 The Arabic question, answered as far as the evidence allows — and it is a "no" from Amazon, not a gap Amazon has filled

Four separate first-party facts, none of which settles the question alone and which together point one way.

(a) Amazon tells sellers the UAE store does not need Arabic. On its own global-selling page for the Middle East [A2-S25]:

"Our UAE store is also an English-language experience, which makes it an opportunity to reach international customers without having to update products, packaging, or services for translation."

This is the strongest single explanation found for why no Arabic listing generation exists anywhere in Amazon's toolset. Amazon's own merchandising position is that it is unnecessary. It also explains 006's parked gap — Arabic listing generation absent from all 100 practitioner sources — the market has been told it does not need it. Note the scope: Amazon makes this claim about the UAE only. It makes no equivalent English-language claim about amazon.sa, whose consumer storefront is Arabic-first. If the Arabic differentiator has an honest home, it is KSA, not the UAE. That distinction is new and belongs in the Brief.

(b) Seller Central UAE has no Arabic web UI. Its own language selector offers exactly two options: 中文 - CN and English - AE [A2-S14]. (For comparison, sellercentral.amazon.com offers 14 languages — Arabic in none of them [A2-S15].) Amazon does ship an Arabic interface on the Amazon Seller mobile app, announced on the .ae forum ≈2023 [A2-S17], but that is interface language, explicitly, and says nothing about generated content.

(c) The asymmetry is documented and quantified. advertising.amazon.com publishes 33,597 URLs across 22 locale prefixes; ar-ae is a full locale with 1,456 pages, and there is no en-ae and no Saudi locale at all [A2-S43]. So Amazon's acquisition funnel for UAE advertisers is Arabic, while the operating surface a UAE seller works in is not. That is a real, dated, first-party-documented gap.

(d) The only explicit language scope Amazon has ever published on a gen-AI tool is English. The US A+ Content announcement: "These tools are available to you at no cost for select English language modules" [A2-S23]. The 2026 EU/CA/IN version drops the qualifier to "select modules" [A2-S24], which is a change of wording, not evidence of Arabic. The canvas roadmap line is future tense and names no language: "Later this year, the canvas experience will begin rolling out to additional countries and will become accessible in languages beyond English" [A2-S4].

What is NOT established, and must not be claimed either way: whether Enhance My Listing generates Arabic text on a live .ae ASIN. The Arabic .ae forum post announcing the tool [A2-S14] is a sentence-by-sentence translation of the English announcement [A2-S15], posted the same day by the same News_Amazon account, and makes no language claim about the output — it is evidence of a localised announcement, not an Arabic product. (Engagement asymmetry worth noting on its own: 774 views / 12 replies on .com, 6 views / 0 replies on .ae.) Parked as P1; resolvable by Ilia in one logged-in session.

And the load-bearing unknown is one level up from that. Nothing in this Area establishes that Arabic listing content converts on amazon.ae or amazon.sa. Amazon's own copy argues the opposite for the UAE [A2-S25]. The strongest honest formulation available is: Amazon says the UAE store is English; the shopper storefront nevertheless offers Arabic; nobody has shown which side is right. Parked as P11 and handed to A5/A3 — it is a demand question, not a platform question.

1.5 Amazon Ads AI: the Gulf is outside four published country lists

The single most useful page in this Area is Amazon Ads' Creative Solutions page, because Amazon footnotes each AI creative tool with an explicit country list [A2-S32]:

¹ (creative partner / Creative Agent) "Available to advertisers in Canada, China, France, Germany, Italy, Spain, United Kingdom, and United States." ² (image generator) "Available in Australia, Belgium, Canada, France, Germany, India, Italy, Japan, Mexico, Netherlands, Poland, Spain, Sweden, United Kingdom, and United States." ³ (video generator) "Available for Sponsored Brands video under Creative Studio in the following countries: Australia, Canada, France, Germany, India, Italy, Japan, Mexico, Spain, United Kingdom, and United States." ⁴ (audio generator) "Available to advertisers in the United States."

Neither the United Arab Emirates nor Saudi Arabia appears in any of the four. And this is a demonstrated absence, not a reading of silence: the Arabic-UAE localisation of the same page carries the identical lists in Arabic, with the UAE still absent [A2-S33]. That is Amazon telling UAE advertisers, in Arabic, on its UAE-locale site, that its AI creative suite runs in a list of countries which does not include theirs.

Two corroborating first-party dates: Creative Agent's own unBoxed launch says "Where is the feature available? North America: United States" while being open to "Vendors · Registered sellers · Managed-service advertisers · self-service advertisers" [A2-S30] — i.e. it is small-seller-reachable, just not in the Gulf. Creative Studio itself launched US-only on 2024-10-15 [A2-S36]. Video Generator for Streaming TV is US-only as of 2025-11-20 [A2-S38]. New Product Campaigns is US-only and category-restricted [A2-S39].

All of these are free. Amazon's wording across the set: "at no cost", "at no additional cost", "all at no cost" [A2-S30] [A2-S35] [A2-S37]. So the entire free AI-ad-creative layer a US seller receives is, on Amazon's own published lists, absent for a Gulf seller. That is the clearest open lane this Area found.

Two Gulf-positive exceptions on the ads side, both real and both narrow. Brand+ (DSP predictive AI) launched globally 2025-01-30 and Amazon published UAE beta results on 2026-06-10 [A2-S40] [A2-S41] — but it is a DSP media product, not a seller tool. And the marketplace-availability widgets on the Sponsored Products, AMC and Amazon DSP product pages flag EG, KSA and UAE available [A2-S42] — the substrate is there, which is what makes the missing creative layer conspicuous.

1.6 Ads Agent and Campaign Manager: available on paper, unreachable for the beachhead

Ads Agent is free — "How much does Ads Agent cost? Ads Agent is available at no cost." [A2-S26] — and its structured marketplace widget flags KSA available, UAE available [A2-S26]. Its human-in-the-loop design is explicit and worth quoting back at anyone who thinks watch/advise/draft is a concession:

"Ads Agent summarizes all proposed changes before taking action, leaving you in control of what changes are made. After your review and approval, Ads Agent makes the specified changes." [A2-S26]

But the eligibility gate makes it unreachable for the Inquiry's beachhead. Amazon's own FAQ: "advertisers with access to Amazon Marketing Cloud and Multimedia Solutions with Amazon DSP are eligible… Availability varies by locale. Reach out to your Amazon Ads account executive" [A2-S26]. The unBoxed launch page is narrower still: "All users with access to Amazon Marketing Cloud (AMC) and U.S. users of Multimedia Solutions with Amazon DSP", accessed via "Amazon Marketing Cloud (AMC) · Amazon DSP" [A2-S27]. A one-SKU amazon.ae seller running Sponsored Products has no AMC seat, no DSP seat and no Amazon Ads account executive, and therefore cannot reach Ads Agent at all.

Campaign Manager (beta) is the same shape: UAE and Saudi Arabia are inside the announced footprint, the persona in the announcement is explicitly a small Sponsored-Products-and-Brands advertiser, and access is "currently available to a select group of advertisers and partners… please reach out to your Amazon Ads executive" with a "broader rollout… later this year" promised on 2025-11-10 [A2-S31]. Whether that rollout has landed in the Gulf is not established (§4 P4).

1.7 The Ads MCP Server: the one first-party surface built for third-party agents

Announced 2026-02-02 by Amazon Ads' VP of Ads Measurement [A2-S34]. This resolves A1's parked item — the vendor's date was right and only the URL was wrong (§3). Verbatim:

"Advertisers and partners can use a single integration to connect custom-built agents or AI platforms such as Claude, ChatGPT, or Gemini."

"…allows agents to access individual Amazon Ads capabilities such as creating, updating, or deleting campaigns; running performance and reporting queries; managing account-level settings; and accessing billing and financial data."

"The Amazon Ads MCP Server and tools are available globally in open beta to Amazon Ads partners with active API credentials."

Three product-story readings. (a) Amazon names Claude explicitly — it is courting third-party agent builders on this surface, not excluding them. (b) "Available globally" is the strongest availability language in this Area and stands in deliberate contrast to the country-listed creative tools. (c) Even Amazon's own flagship agentic tool stops where SCOPE puts v1: the end-to-end campaign tool "generates a ready-to-launch campaign that only needs review and approval" [A2-S34].

Access mechanics belong to A6 and are not restated here; A6 established the prerequisite is an LwA application with Ads API access, self-applicable by a Direct Advertiser in up to one business day, served to AE and SA from advertising-ai-eu.amazon.com/mcp A6:[A6-S24].

1.8 noon ships one AI feature for sellers, and it is a category classifier

After fetching and full-text-searching all 243 English help-centre articles (1,462,478 characters) [A2-S64], noon's entire published seller-facing AI surface is one button in Seller Lab's listing flow, described in two articles and one paragraph:

"Product Category: Click Generate Product Category to automatically assign the best classification for your item." [A2-S65]

"AI-Generated: Choose from the smart suggestions provided based on your product info. Select Manually: Browse and pick the category hierarchy yourself if the AI suggestions don't fit." [A2-S66]

There is no seller assistant, no chatbot, no listing-copy generation, no translation, no forecasting assistant and no AI-branded ad automation in noon's documentation. And it is not exposed on the API — noon's machine-readable API index lists only ListCategories, ListCategoryAttributes and GetExportCategoryList, with no category-prediction, content-generation or translation endpoint [A2-S74]. The classifier is a UI convenience, not a programmable capability.

The absence was demonstrated, with a stated limit. Term frequency across all 243 articles [A2-S64]: artificial intelligence 0 files, AI-powered 0, machine learning 0, generative 0, chatbot 0, copilot 0, smart pricing 0, recommendation engine 0; standalone token AI appears in 2 files, both the category paragraph above. A URL-slug sweep of all 488 sitemap URLs for eleven AI-adjacent stems returned four articles, all false positives (automated reimbursement, automated return preference, two "recommended RTV report" pages) [A2-S63] [A2-S64].

The limit, stated honestly: every Seller Lab path returns HTTP 307 to login.noon.partners with header x-authproxy: l9, and — unlike Amazon's Appstore — there is no publicly fetchable Seller Lab front-end bundle to inspect [A2-S75]. Anything noon ships inside Seller Lab that it has not documented is invisible to this research. The correct claim is therefore: "noon documents no seller AI beyond category suggestion," not "noon has no seller AI."

1.9 noon's written refusal to translate — the cleanest wedge in this Area

This is a policy, published on noon's own help centre, in a Q&A [A2-S67]:

"Is noon providing the translation for the content of the Seller SKU?" "We are not providing translations for the Seller SKU products. Every seller is responsible to create the content in both languages if needed."

With a stated commercial consequence on the same page:

"If you only provide one language content the visibility of your SKU might be limited. For example, if you only provide English content for a product being sold on the UAE website, that product might only get shown on the English website and be hidden from the Arabic one."

The language matrix, verbatim [A2-S67]:

The operational surround corroborates it: titles are to be provided "in both English and Arabic" [A2-S65]; content status is tracked separately per language [A2-S69]; the bulk listing flow offers an English-only or English+Arabic template with "we strongly recommend… English and Arabic" [A2-S70]; brand names must be transliterated into Arabic, with wrong spelling and N/A both listed as rejection causes [A2-S71]; and the closest thing noon offers to help is a CSV round-trip — "download, modify, and update all content (in English and Arabic)" [A2-S68].

So: noon requires Arabic for Arabic-site visibility, rejects bad Arabic brand transliteration, hands you an empty Arabic column in a spreadsheet, and says in writing that it will not fill it. On the Amazon side, only 83 of 1,386 AE-capable Appstore apps declare Arabic support at all (§1.12). Nobody is filling this. Note the honest counterweight from §1.4: on noon the Arabic requirement is documented and commercial; on Amazon.ae the platform's own position is that Arabic is unnecessary. The Arabic case is strongest on noon and on amazon.sa, and weakest exactly where Ilia sells today.

1.10 What noon ships that is not AI but occupies adjacent ground

Two first-party noon products matter for foreclosure even though neither is AI-branded.

Vantage Analytics [A2-S72] — "a powerful analytics platform designed specifically for products to gain granular, data-driven insights into sales, customer behavior, and market trends", with modules for Purchases (sales performance, funnel), Customers (segment, loyalty), Partner Rank ("monitor your competitive position"), Search ("Product Search Ranking and Top and Search Terms insights") and Commercial (pricing, stock, assortment). This is a first-party, Helium-10-shaped analytics layer already inside Seller Lab — including the search-term data A6 found has no API equivalent A6:[A6-S32]. But it is gated: "If you cannot see the tool or need access enabled, please reach out to your Account Manager or email seller@noon.com" [A2-S72]. An account-manager gate means a brand-new solo importer almost certainly does not have it. noon has built the analytics and has not made them self-serve.

noon ads [A2-S73] — "a suite of self-serve advertising solutions" with an Ad Manager. The help centre says nothing about automated bidding, AI budget rules or campaign generation, and advertise.noon.com was unreachable from this session (HTTP 000, plausibly Akamai geo-blocking) [A2-S81]. Parked, not concluded (§4 P12). This is consistent with A6, which found no advertising API in noon's 51 documented endpoints A6:[A6-S32] — noon sells ads in the UI and does not expose them programmatically.

And noon forecloses the automation workaround contractually. Its UAE, KSA and Egypt seller agreements, all published 2026-04-22, carry an identical clause 8.3: "You must not use any automated tool of any type to scrape data from the noon platform. Such activity will be considered to be a material breach of the Agreement" — and clause 7.5 makes clause 8 survive termination [A2-S82]. Combined with the fully walled Seller Lab [A2-S75], browser automation of noon Seller Lab is a contractual breach, not a grey area. The sanctioned path is the partner API A6 documented. (One documented gap: the International Seller T&Cs contain no clause 8 and no anti-scraping clause at all [A2-S82] — recorded, not relied upon.)

1.11 Partner posture, part 1: Amazon is courting AI assistants, in writing

This materially qualifies the 006:[L09]-era framing that Amazon is hostile to AI agents.

sell.amazon.com/ai-information, page stamped "Last updated: August 2026" [A2-S44]:

"This page provides structured information about Sell on Amazon, intended for AI assistants such as ChatGPT, Claude, Perplexity, Gemini, and other large language models (LLMs)."

The same page publishes Amazon's own inventory of its seller AI — Seller Assistant, Product Opportunity Explorer, Enhance My Listing, A+ Content Generator, AI Video Ad Creator and Creative Agent [A2-S44]. And the Ads MCP Server names Claude, ChatGPT and Gemini as intended clients [A2-S34].

There are two Amazon postures, and 006 observed only one of them. The hostility 006 recorded — crawler blocks August 2025, the Perplexity suit November 2025 006:[L09] — is about the shopper-facing storefront, and A6 confirmed it is still in force and wider (~101 blocked user-agents on amazon.ae/robots.txt and amazon.sa/robots.txt, ClaudeBot and GPTBot included) A6:[A6-S28]. The seller and developer surface is a different door, and Amazon is holding it open on purpose.

Demonstrated absence check on the GCC side of that welcome: sell.amazon.com publishes 391 sitemap URLs including /ai-information and a full /tools/* section [A2-S46]; sell.amazon.ae publishes 111 URLs, of which zero mention AI, generative or artificial, and there is no /tools/ section at all [A2-S45]. sell.amazon.ae/tools returns HTTP 404 [A2-S11]. The .ae and .sa seller-site footers carry no Tools section, no Brand Name Generator and no Seller University, where the US has all three [A2-S13]. Amazon does not market any AI seller tooling to UAE sellers on its own UAE seller-acquisition site.

1.12 Partner posture, part 2: and regulating them — the economics, the cap, and the clause

Amazon welcomes third-party seller apps on terms that are, economically, unusually generous, and contractually, unusually specific about agents.

No listing fee, no revenue share — proven from the governing contract, in the UAE locale. A full-text search of the Solution Provider Portal Agreement, Version August 2026, en_AE (70,029 characters) returns commission 0, revenue share 0, percent 0, % 0; the single royalt hit is "royalty-free" in the API licence grant, and all 18 fee hits are arbitration fees, attorneys' fees, taxes, or a conditional placeholder for other paid programmes [A2-S56]. Amazon monetises the seller, not the app vendor — materially different from Shopify's 15–20 % or the mobile app stores. Amazon's own FAQ confirms the vendor sets price and owns support: "Software partners set their own prices"; "Amazon is not directly involved in the development or sale of third-party apps" [A2-S47].

Listing is effectively mandatory, and the number is 25. The docs hedge ("Although you're not required to list your app…" [A2-S52]), the contract does not ("we require you to list your Application in the SPA if you sell your Application or offer it for use to Solution Provider Customers" — SPPA D-1 [A2-S56]), and the authorization-limits table makes the argument moot [A2-S54]:

A seller-facing SP-API app that is not listed in the Appstore hits a hard ceiling at 25 authorised selling partners. Getting to customer #26 requires passing the Appstore review, and Amazon states publication takes "three to four weeks" after approval [A2-S52]. Review prerequisites include a live public website that is not under construction and not login-only [A2-S53]; PII-touching roles add an architecture review with the SP-API Solutions Architecture team, "a detailed explanation of data flows and protection controls for PII" that "can involve a demo through screen sharing" [A2-S53]; applicants must respond within five days or the case closes [A2-S53]. Amazon reserves explicit discretion to refuse (SPPA D-3) and to delist with 30 days' notice, or immediately (D-4) [A2-S56].

Amazon's agent regime is written into the UAE developer contract. SPPA en_AE, Version August 2026 [A2-S56]:

"1.3 Use of Agents. If you deploy, enable, or authorize an Agent to access or interact with any Amazon Services… a. Identification and Compliance Requirements. All Agents must clearly identify themselves as automated systems… Agents are prohibited from accessing Amazon Services if Amazon has explicitly requested cessation of such access. b. Access Limitations. Amazon reserves the right, at its sole discretion, to limit or restrict Agent access… c. Technical Compliance Standards. Technical compliance standards are set forth in the corresponding Agent Policy."

And the definition, from Appendix 1:

"'Agent' means any software or service that takes autonomous or semi-autonomous action on behalf of, or at the instruction of, any person or entity."

That definition captures the entire product concept in SCOPE, including the "watch, advise, draft" v1. This partially closes A6's open item (§3): A6 could not obtain the Agent Policy or BSA §19 text, and correctly parked it; A2 can now show that the developer-side agreement a UAE-registered solution provider signs independently carries the agent regime, in the en_AE locale, with the same definition A1 read from Amazon's seller-side announcement A1:[A1-S32]. The agent regime binds the vendor as well as the seller. The Agent Policy document itself remains unread (§4 P13).

Four AUP clauses foreclose specific product ideas outright [A2-S58] — all confirmed by A6 from the en_US text A6:[A6-S7], and reconfirmed here in the en_AE locale:

• 4.4 — "Do not aggregate data across Authorized Users' businesses or customers obtained through the Amazon Services API to provide or sell to any parties, including competing Authorized Users." → cross-seller benchmarking is prohibited. "Sellers like you price at X" is not buildable on SP-API data.
• 4.3 — "Do not use, offer or promote external (non-Amazon) data services that vend information or data retrieved from Amazon's websites." → reselling scraped market data inside the app is a policy problem.
• 3.6 / 3.7 — do not ask users to hand over portal data to circumvent policy; if portal access is needed, "ask the Authorized User to grant access through secondary user permissions". → the sanctioned alternative to automating a seller's login is a secondary user account.
• 2.4 / 2.10 — "Be explicit about any calculations and the use of models such as artificial intelligence… their accuracy and data freshness" and "Implement data integrity and validation checks… for any analytical processing (e.g. AI models for insights, automated decision-making) that has material impact on an Authorized User's business." → AI disclosure and validation are contractual obligations, not product choices.

One Gulf-specific operational detail: AUP §5.1 lists "United Arab Emirates Communication Guidelines" and "Kingdom of Saudi Arabia Communication Guidelines" as separate documents [A2-S58]. An AI that drafts buyer messages needs the AE and SA variants, not the US one.

And the compliance floor rose twelve days before this Inquiry ran. Amazon's own announcement, effective 2026-08-25 [A2-S60]:

"…we're updating the agreement and policies that govern how you use the Solution Provider Portal to strengthen security standards. These changes apply to all solution providers… The Data Protection Policy now requires a web application firewall for internet-facing endpoints, storage encryption on all devices, a 15-minute screen lock, and a 30-minute lockout after ten failed login attempts. You must also designate an Incident Management Point of Contact, encrypt all credentials at rest, delete information within 30 days when required, and meet expanded vulnerability scanning and penetration testing* requirements…"

"…continued use of the Selling Partner API (SP-API) after August 25 means you accept the updated agreement and policies."

Confirmed live in the UAE locale: the en_AE Data Protection Policy already carries the WAF requirement, the 15-minute screen-lock, MDM/UEM for unmanaged devices, penetration testing "at least every 365 days", code scanning "prior to each software release", the named IMPOC and 30-day deletion [A2-S59].

For a two-person GCC startup this is a datable line item, not a wall — MDM on every laptop, a WAF, an annual third-party pen test, per-release code scanning, a named incident contact and possibly insurance. It lands before customer #26, not after. An accelerator pitch should carry it as a known cost.

One more obligation that lands directly on any AI content product, announced ≈July 2026 and scoped globally [A2-S61]:

"You must now tag product images that contain photorealistic AI-generated people with specific metadata before you add them to your listings and A+ content in any of our worldwide stores… you must use an IPTC-compatible metadata editor to add the keyword contains-synthetic-performer to the dc:subject (XMP) field."

"In any of our worldwide stores" means amazon.ae and amazon.sa. Any product generating lifestyle imagery with people for Gulf listings inherits this XMP-tagging obligation. (Third-party reporting attributes the rule to New York synthetic-performer disclosure legislation — Forbes, 2026-07-25, secondary [A2-S62].)

1.13 The Appstore, read as a market map

The Selling Partner Appstore turns out to have a public, unauthenticated, queryable API that answers on sellercentral.amazon.ae — no account, no cookie, no CSRF token [A2-S48] [A2-S49] [A2-S50]. That makes "which marketplaces does tool X support?" answerable from Amazon's own data rather than from vendor marketing, for any tool. Method and gotchas are in [A2-S50]; the numbers below were all read on 2026-09-06.

Reach. 3,638 apps globally. 1,386 declare support for AE (49.5 % of the 2,802 that declare US), 1,305 for SA, 1,211 for EG [A2-S50]. A UAE seller sees roughly half the catalogue a US seller sees — a real long tail, and a much better position than "the Appstore does not serve the Gulf."

Arabic. marketplacesIds=AE&supportedLanguages=ar → 83. That is 6.0 % of AE-capable apps declaring Arabic, against 1,117 declaring English (80.6 %) and 582 declaring Chinese (42.0 %) [A2-S50]. Chinese support is seven times more common than Arabic in the store a UAE seller browses. Caveat, stated because it matters: supportedLanguages is self-declared and most plausibly means the app's own interface or support staffing, not generated-content language. It is a ceiling on Arabic capability, not a measure of it (§4 P10). Of those 83, roughly a quarter are Chinese cross-border ERPs that tick every language box; fewer than ten are genuinely regional (mazeed/McLedger, Nasam, Limonene, Zid Marketplace Connect, Rabta, OTO, Omniful, SouqMetrics, Dunes & Deals). Zero of the 83 is a guided end-to-end onboarding product.

The taxonomy is the finding. The Appstore has 16 categories in 7 groups. For AE the largest are Inventory and Order Management (912), Listing (731), Analytics and Reporting (720), Shipping Solutions (413), Advertising Optimization (358) [A2-S50]. There is no "AI" category — and no Compliance, Sourcing, Freight, Customs, Company Formation or Training category. Every category in Amazon's taxonomy presumes you already have a product and a live account. That is the same absence A1 found in the product landscape, expressed as a data model.

The nearest incumbent, and the proof of the gap in one record. Jarvio — "your AI-powered e-commerce expert… launch, manage, grow and understand your brand, without the learning curve", with a "Jarvio Assistant" and a "Listing Builder" — is live on .ae, .sa and .eg, across 24 stores, at USD 149/mo, declares Arabic among 17 languages, and has zero Appstore reviews. Its own usageScenario field reads: "Brands. Agencies serving Brands." [A2-S51]. The closest existing thing to the SCOPE concept is already in the Gulf, is priced and targeted at brands and agencies rather than first-time importers, and nobody has reviewed it.

Two further primary corrections to A1's competitive set, from Amazon's own data [A2-S51]: Jungle Scout and ZonGuru declare no Gulf marketplace at all (10 stores each, AE/SA/EG absent), while Helium 10 declares AE, SA and EG across 21 stores — settling A1's parked "Helium 10 GCC coverage" item from a primary source (§3). Review volume across the whole set is tiny (Helium 10: 16; sellerboard: 15; most: 0) — Appstore ratings are neither a moat nor a discovery signal.

The .ae Appstore marketing page is a US page in Gulf clothing. Every CTA on sell.amazon.ae/programs/selling-partner-appstore points at sellercentral.amazon.com with sdus (US campaign) tracking tags; the developer route links to developer.amazonservices.com with a US tag; the footer reads "© 2025 Amazon.com, Inc."; the language selector offers English-US and Español-US and no Arabic [A2-S47]. Meanwhile Amazon's developer paperwork is Arabic-localised for the UAE — a full Arabic SPPA exists at ?locale=ar_AE [A2-S57], with the standard "the English version will prevail" rider [A2-S55]. Amazon localises the contract and not the storefront.

1.14 The clock

Amazon Accelerate 2026: 22–24 September 2026, Seattle Convention Center [A2-S12] — 16 days after this Finding. Accelerate is Amazon's habitual venue for seller-AI announcements: gen-AI listing creation was announced at Accelerate 2023 [A2-S18], Enhance My Listing and Project Amelia at Accelerate 2024 [A2-S19], the agentic Seller Assistant and Creative Studio's Creative Agent at Accelerate 2025 [A2-S1] [A2-S16]. Mary Beth Westmoreland — the author of the Seller Assistant press release — is a named featured speaker, and the forum post promises "general sessions with exclusive announcements" and an "Innovation Hub where you can test new Amazon tools" [A2-S12].

Recommendation for the Brief: carry an explicit "re-check A2 after 2026-09-24" flag. A .ae/.sa Seller Assistant rollout announced at Accelerate 2026 would materially change §1.2 and §2.1 — though note that it would not change §1.2's scope disclaimer, which is the part of the foreclosure argument that does not depend on geography.

---

2. Capability tables

2.1 Amazon first-party seller AI — feature inventory and store availability

Legend: Y = documented available on a first-party Amazon page on 2026-09-06 · N = documented not in Amazon's published availability list · ? = no first-party country statement found (see §4; a "?" is not a "no") · gate = available in the store but behind an eligibility wall.

Method for rows 1, 1b and 2. Amazon's help hub serves the Seller Assistant article GLYJTRGQCYNYAZS9 at a public per-store URL. It was fetched on nine Seller Central hosts through a JS-rendering proxy, and each host was controlled with article GXMJ38VA95GUN5XU ("New Seller Incentives", linked from the .ae footer) [A2-S13]:

Help-article presence is a strong proxy, not proof — see §3 and §4 P2.

2.2 noon first-party seller AI — the inventory, and where I looked

Term frequency across all 243 English help-centre articles (case-insensitive file counts) [A2-S64]:

Where else I looked and found nothing: noon's partner-API machine index [A2-S74]; a URL-slug sweep of all 488 sitemap URLs for eleven AI stems, yielding four false positives [A2-S63]; and the seller hostname map [A2-S81]. Where I was blocked: Seller Lab itself (HTTP 307 to login on every path, no public bundle) [A2-S75], advertise.noon.com (HTTP 000) [A2-S81], training.noon.partners (login-gated) [A2-S79], and noon's corporate newsroom (not swept). See §4.

2.3 The Selling Partner Appstore as a Gulf market map (Amazon's own facet data, 2026-09-06)

Selected per-app store coverage, from Amazon's own detail endpoint [A2-S51] — this settles several items A1 had to park on vendor silence:

---

3. Reconciliation notes

Reconciliation note — A1's Seller Assistant contradiction. Dissolved; both observations were true. A1:[A1-S38] (a competitor blog) described a free, always-on, agentic Bedrock/Nova/Claude assistant for third-party sellers; A1:[A1-S35] (sell.amazon.com/tools) lists no such product and only an "AI brand name generator". The two describe two different surfaces: sell.amazon.* is Amazon's pre-login acquisition-marketing site and its tools index catalogues publicly linkable tools, while Seller Assistant is a post-login Seller Central surface documented in the help hub [A2-S3] and on Amazon's partner site [A2-S2]. Absence from /tools says nothing about whether it ships. Item-by-item against Amazon primary sources: "always-on, agentic AI-powered partner" confirmed verbatim [A2-S1]; Bedrock + Nova + Claude confirmed verbatim [A2-S1]; free to all third-party sellers confirmed twice [A2-S1] [A2-S3]; "US from December 2025" corrected — Amazon's agentic announcement is 2025-09-17 [A2-S1]; "EU/UK through Q1–Q2 2026" broadly confirmed (UK ~March 2026 [A2-S5]); "tools index shows only a brand name generator" confirmed [A2-S10].

Reconciliation note — the "autonomy tiers" claim. Not supported; do not quote it. A1 recorded, from A1:[A1-S38], tiered "suggest only" vs "auto-approve" permissions. No Amazon source found in this Area documents an auto-approve or act-unattended mode. What Amazon documents is (a) a standard Seller Central None / View / Edit user permission named "Seller Assistant", administrator-grantable [A2-S3], and (b) per-action approval inside the conversation ("If the seller approves…", "Once approved…") [A2-S1] [A2-S3]. Resolved in favour of the primary documentation; the tier claim is parked (§4 P3) as a probable blog paraphrase of those two mechanisms.

Reconciliation note — Seller Assistant on .ae: an absent help article versus a .ae page that names the product. An Amazon-published, .ae-hosted forum post dated 2026-06-15 says a new Add Products search bar lets sellers "access Seller Assistant without leaving the page" [A2-S8]. Read alone that implies .ae availability. But the identical post — same title, same body — is published on the US and EU forums with only en-GB spelling changes; it is a globally syndicated News_Amazon feature post describing the global Add Products redesign [A2-S8]. Resolved in favour of absence, on three grounds: the help article is missing on .ae and .sa against a control that renders on the same host the same day [A2-S13]; Amazon's own and only availability sentence is "all sellers in the U.S. store… other countries in the coming months" [A2-S1]; and a DuckDuckGo site:sellercentral.amazon.ae "seller assistant" probe returned exactly one hit — that same syndicated post. Confidence: high, not proven. The countervailing post is flagged rather than buried, and §4 P2 names the 60-second test that settles it.

Reconciliation note — Ads Agent in UAE/KSA: Amazon contradicts Amazon. The Ads Agent product page's structured marketplace-availability widget flags KSA and UAE available with no skill qualifier [A2-S26]; the unBoxed launch page lists the same Middle East footprint but gates users to "All users with access to AMC and U.S. users of Multimedia Solutions with Amazon DSP" [A2-S27]; and Amazon's UAE-domain help article, updated 2026-04-15, says "Using Ads Agent for optimizing campaigns and campaign targeting is currently limited to advertisers in certain countries across North and South America" [A2-S28]. Not fully resolved. The most probable reading — AMC/media-plan skills global, optimisation and targeting US-first — is recorded as a reading, not a finding, because Amazon never names the "certain countries" (§4 P5). What is resolved and is the load-bearing point either way: the gate is AMC + Amazon DSP + an account executive, which the Inquiry's beachhead does not have.

Reconciliation note — the Arabic .ae announcement of Enhance My Listing. A3's parked item [A3-W21] asked whether the Arabic .ae forum post [A2-S14] proves Amazon generates Arabic listing text. It does not. The Arabic post is a sentence-by-sentence translation of the English post [A2-S15], same author account, same timestamp, with the product name left in Latin script and no statement about output language in either version. Resolved: it is evidence of a localised announcement, not of an Arabic product. The nearest language evidence points the other way (§1.4). Parked as P1.

Reconciliation note — Amazon's posture toward AI, and 006:[L09]. 006 recorded Amazon as hostile to AI agents (crawler blocks Aug 2025, the Perplexity suit Nov 2025). That remains true of the shopper-facing storefront — A6 confirmed ~101 blocked user-agents including ClaudeBot and GPTBot on amazon.ae/robots.txt and amazon.sa/robots.txt A6:[A6-S28]. It is not true of the seller and developer surface. Amazon publishes an LLM-addressed page naming Claude [A2-S44], ships an MCP server naming Claude, ChatGPT and Gemini [A2-S34], charges no listing fee and no revenue share [A2-S56], and localises its developer contract into Arabic [A2-S57]. Resolved: two doors, two postures. The correct framing for the Brief is regulated welcome, not hostility.

Reconciliation note — the Amazon Ads MCP 404. A1 parked a vendor claim that an Amazon Ads MCP server launched 2026-02-02 while advertising.amazon.com/blog/amazon-ads-mcp-server 404s. Both halves check out: the URL genuinely returns HTTP 404 (verified 2026-09-06), and the real article exists at /library/news/amazon-ads-mcp-server-open-beta, authored by Amazon Ads' VP of Ads Measurement and dated February 2, 2026 [A2-S34]. The vendor's date was correct; only the URL was wrong. A1's parked item and 006:[L09]'s "partner-only" framing are both now closed — the latter by A6 A6:[A6-S24].

Reconciliation note — the Agent Policy, and what A2 can add to A6's parked item. A6 could not retrieve BSA §19 or the Agent Policy text (the Seller Central content API returns {"reason":"sign_in"}) and correctly parked it A6:[A6-S26]. A2 did not retrieve that document either (§4 P13) — but it did retrieve the developer-side equivalent in the UAE locale: the Solution Provider Portal Agreement, Version August 2026, en_AE, carries §1.3 Use of Agents and the Appendix 1 definition "any software or service that takes autonomous or semi-autonomous action on behalf of, or at the instruction of, any person or entity" [A2-S56] — the same definition A1 read from Amazon's seller-side announcement A1:[A1-S32]. Resolved to this extent: the agent regime binds the vendor as well as the seller, and it is in the contract a UAE-registered solution provider signs. The Agent Policy's technical standards remain unread and every specific number circulating in trade blogs (12-month audit trails, 180-day logs, a 20 % price-change threshold, a certification programme) remains unverified, exactly as A6 concluded.

Reconciliation note — noon advertising: no contradiction with A6. A6 found no advertising or sponsored-products API among noon's 51 documented endpoints and parked whether an undocumented one exists A6:[A6-S32]. A2's help-centre sweep is consistent: noon ads exists as a self-serve UI product with an Ad Manager [A2-S73] and no documented automation, and it is not on the API. The two findings describe the same shape — noon sells ads in the console and does not expose them programmatically. advertise.noon.com was unreachable from this session, so neither Finding may claim absence of AI in noon ads (§4 P12).

Reconciliation note — handed to A4: noon's Seller Lab training is not mandatory. A4 carried a third-party claim that noon requires new sellers to complete onboarding training. Against noon's own pages: "monthly" is confirmed ("Request the monthly training calendar at seller@noon.com" [A2-S76]); "English and Arabic" is confirmed verbatim ("Q. In which languages are the training sessions available? A. The sessions are available in English and Arabic." [A2-S77]); "required" is not confirmed and the primary pages contradict the framing — every noon page presents training as free, recommended and bookable, training appears in only 4 of 243 articles, none using required, must attend or mandatory, and the seller-eligibility pages do not mention training at all [A2-S64]. What noon does do is sequence it: training is step 1 of the post-approval article, before "list your products" [A2-S76], which is the most likely origin of the third-party claim. Recommend A4 downgrade to: noon presents onboarding training as step 1 for a newly approved seller, free, monthly and bookable in English or Arabic, but does not condition selling on it.

Reconciliation note — A1's parked GCC-coverage items, settled from Amazon's own data. A1 could not read Helium 10's or Jungle Scout's help centres (Cloudflare challenge) and parked their Gulf coverage; it also parked a dozen vendors whose marketing pages are silent on .ae and .sa. The Appstore detail endpoint answers this from Amazon's data rather than the vendor's [A2-S51]: Helium 10 declares AE, SA and EG across 21 stores; Jungle Scout and ZonGuru declare no Gulf marketplace at all (10 stores each) — the latter corroborating ZonGuru's own pricing-page statement A1:[A1-S24]; and Perpetua, Teikametrics, sellerboard, Superfuel, Trellis (AE only) and Sellerise all declare AE. Caveat that keeps this honest: marketplaceIds is what the vendor declared on its Appstore listing, not a tested integration, and it says nothing about feature-level parity inside the tool. It resolves "is it offered here", not "does everything work here".

---

4. Open Questions / Parked

Method limitation, stated up front so it is never mistaken for a finding. The session's WebSearch budget was exhausted before all three A2 research passes ran. Every discovery below the Amazon and noon domains used curl, WebFetch against known URLs, sitemap.xml and robots.txt enumeration, PyPI/GitHub APIs, and DuckDuckGo-lite through r.jina.ai. Where this Finding says "not found", read "not found by targeted fetch or sitemap enumeration" — except where a probe method with a control is stated (§2.1 for Seller Assistant, §2.2 for noon), which is the only place demonstrated absence is claimed. scrapegraph was unavailable all session (SGAI_API_KEY missing). Additional specific blocks are named per item below.

P1 — Does Enhance My Listing generate ARABIC listing text on amazon.ae? UNRESOLVED, and it cannot be resolved from public documents. Every public artefact — the Arabic .ae forum post [A2-S14], the English .com post [A2-S15], the About Amazon launch article [A2-S19], and the .com/.ae/.sa help article [A2-S20] [A2-S21] [A2-S22] — describes the tool identically and says nothing about output language. Nearest evidence points the other way: A+ gen-AI is scoped to "select English language modules" [A2-S23], and Seller Central UAE offers no Arabic UI [A2-S14]. What settles it: Ilia opening SHIO's live .ae ASIN → Edit listing → Product details → View enhancements, and reading the script of the suggestions. SCOPE assigns SHIO account state to Ilia, not to a Scout. Do not let the Brief claim Amazon has closed the Arabic gap, and do not claim it definitely has not.

P2 — Definitive Seller Assistant availability on .ae/.sa. Help-article presence is a strong proxy (Amazon publishes the article in the stores where the feature is live; the UK article appeared alongside the UK launch), and the control experiment rules out a render artefact [A2-S13]. It is not the same as logging in. Confidence: high but not proven. Countervailing evidence is the syndicated .ae forum post [A2-S8] (§3). The 60-second definitive test, which Ilia can run: Seller Central .ae → Settings → User Permissions → search the permission list for an entry named "Seller Assistant" [A2-S3]; and open Add Products on .ae and check whether the in-page search bar offers Seller Assistant or only help topics [A2-S8]. A screenshot of either settles the biggest platform-risk question in the Inquiry.

P3 — Seller Assistant autonomy tiers. A1's "suggest only vs auto-approve" A1:[A1-S38] is not found in any Amazon source (§3). Parked as unverified; do not quote it.

P4 — Did Campaign Manager's "broader rollout… later this year" reach .ae/.sa? Promised 2025-11-10 [A2-S31]. No later Amazon page confirming general availability was found, in the Gulf or anywhere. Not determined.

P5 — The exact split of Ads Agent skills by country. Amazon never names the "certain countries across North and South America" in [A2-S28]. The AMC-global / optimisation-US-first reading (§3) is the most probable but is a reading. Parked as a stated ambiguity.

P6 — Country availability of gen-AI listing creation (row 3) and A+ generation (row 5) on .ae/.sa. No first-party country list exists for either. For A+, no sellercentral.amazon.ae forum announcement was found — but the .ae forum is thinly indexed by public search engines and its own search endpoint requires JS/auth, so this is a failure to find, not a demonstrated absence. Do not infer these rows from row 4 (Enhance My Listing), which was proven separately.

P7 — Germany launch date (2026-06-29) and the exact UK launch date. Germany rests on a single trade blog that self-describes as reporting more than Amazon published [A2-S7] — parked. The UK date rests on an Amazon forum post that renders a relative timestamp ("6 months ago" on 2026-09-06) [A2-S5] plus a secondary date [A2-S6]. Report as "~March 2026", never as a precise date.

P8 — The 230,000 monthly users and 90 % acceptance figures. Published without a date on Amazon's page [A2-S2]; the 90 % figure is attributed elsewhere to the 2025 Amazon Small Business Empowerment Report [A2-S44]. Cite as "as published, accessed 2026-09-06", never as a dated 2026 metric. Amazon's total 3P seller count was not sourced in this Area — do not use a remembered figure to contextualise 230k.

P9 — Egypt (amazon.eg). Both the Seller Assistant article and the control failed to render on sellercentral.amazon.eg [A2-S13]. Inconclusive, not absence.

P10 — What supportedLanguages means on an Appstore listing. The listing form's adjacent field is labelled "Customer support languages: Select the languages accommodated by your customer support team" [A2-S52], which suggests support staffing; the search facet is supportedLanguages, a different label [A2-S50]. The 83-app Arabic figure is a ceiling on Arabic capability, not a measure of Arabic listing generation. State it that way.

P11 — Does Arabic listing content convert on amazon.ae / amazon.sa? Not addressed by any source in this Area, and Amazon's own copy argues the opposite for the UAE [A2-S25]. This is the load-bearing unknown behind the founder's Arabic differentiator — larger than P1. It needs demand-side evidence (A5/A3), not more platform documentation.

P12 — Whether noon ads ships AI or automated bidding. advertise.noon.com returned HTTP 000 from this session (plausibly Akamai geo-blocking, not absence) [A2-S81]. Do not claim absence.

P13 — The Amazon Agent Policy text itself. SPPA §1.3(c) says "Technical compliance standards are set forth in the corresponding Agent Policy" without hyperlinking it [A2-S56]. It is not in the SP-API docs (developer-docs.amazon.com/sp-api/llms.txt, 906 indexed pages, has no Agent Policy entry; /sp-api/docs/agent-policy 404s), not in the per-locale policies table [A2-S55], and ?documentType=AGENT/AgentPolicy both return HTTP 400. Third-party legal commentary (secondary, unverified) places it as Section 19 of the Amazon Services Business Solutions Agreement, announced 2026-02-17, effective 2026-03-04 — i.e. in the seller's agreement, not the developer's, which would explain its absence from developer docs. Action: fetch it from the Amazon Services Business Solutions Agreement, .ae locale, from a logged-in Seller Central session. That is A6's document set and Ilia has the login.

P14 — The Appstore's "2500+ applications" claim versus the live totalCount of 3,638. Both numbers are proven to exist [A2-S47] [A2-S50]; whether the marketing page is stale (its footer reads "© 2025") or counting something different (distinct providers, a quality bar) is not determined.

P15 — A defensible count of AI-native apps available on .ae. The Appstore keyword matcher is loose (keywords=AI returns 741 of 1,386 AE apps; artificial intelligence returns 208) [A2-S50]. Do not quote 741 or 208 as counts of AI apps. A real number needs enumerating all 1,386 at the 48-item page cap (~29 calls) plus a detail fetch and classification pass per app. Not attempted.

P16 — Whether anything noon ships inside Seller Lab is undocumented. Seller Lab returns HTTP 307 to login on every path with no public front-end bundle to inspect, unlike Amazon's Appstore [A2-S75]. The noon-AI negative rests on noon's own help centre and API docs — strong, but not the same as inspecting the product. Say "noon documents no seller AI beyond category suggestion", not "noon has no seller AI". Related: training.noon.partners is proven to exist with three locale project ids but its catalogue is login-gated [A2-S79]; the three Microsoft Bookings calendars redirect correctly but returned HTTP 417 to this client, so the literal cadence on them is unread [A2-S80]; and noon's corporate newsroom was not swept — a seller-AI press release that never reached the help centre would have been missed.

P17 — sell.noon.com, sellerlab.noon.com, partners.noon.com. All resolve in DNS (wildcard) but returned HTTP 000 over both http and https from this network [A2-S81]. Could be geo/Akamai blocking rather than absence. A6 separately established that noon runs a wildcard DNS zone where non-existent services return HTTP 418 from an Envoy catch-all, so 418 means "does not exist" and 000 means "unreachable from here" A6:[A6-S44]. These are 000s.

P18 — Creative Agent / Video Generator pricing beyond "no cost", and Amazon Ads Academy course content on Creative Agent (sign-in gated) — both unread. No page states a paid tier for any of the free-labelled AI tools; media spend is separate and unquantified.

P19 — Whether an app listed only on sellercentral.amazon.com is installable by a seller logged into sellercentral.amazon.ae. The Appstore API is store-agnostic for reading and marketplaceIds is a per-app field, so the filter is by declared marketplace rather than console domain — but the install/authorize path for a UAE seller was not tested (no seller account was used; anonymous .ae userMetadata returns canUseAppstore:false, which is an install gate) [A2-S49]. Inferred, not proven.

P20 — Amazon's tools index and Seller Central help-hub search API. /help/hub/mons-api/SearchSuggestions returns {"reason":"sign_in"} on every marketplace — a tooling limit, not a finding. Recorded so no one re-burns budget on it. Same for [A2-S29], which was captured only as a search snippet rather than a full page fetch.

---

5. What this does NOT cover

• The technical ceiling. Which SP-API / Ads API / noon API operations are writable, the role model, rate limits, endpoints and marketplace IDs — A6. A2 covers the products Amazon and noon ship to sellers and the contracts a vendor signs, and cross-references A6 rather than restating it.
• The competitive set of third-party tools. A1. A2 contributes only the primary store-coverage data the Appstore detail endpoint yields [A2-S51], which corrects several of A1's parked items — it does not re-survey the category.
• Regional MENA supply-side products and Arabic commerce AI. A3.
• noon's other verticals — noon Food, noon Minutes, NowNow, Namshi, and noon payments. The recovered evidence note establishes these are architecturally disjoint from the marketplace seller surface, that noon Food consumes POS APIs rather than exposing one, and that the corporate relationship between noon payments and noon.com marketplace is not determined. None of it is A2's subject.
• Shopper-facing AI. Rufus, Alexa for Shopping and the storefront crawler blocks are out per SCOPE except where they change seller behaviour; 006's finding that Rufus is unconfirmed on .ae is not revisited here.
• Whether Amazon's or noon's AI is any good. Except for three dated seller replies on Amazon's own forum [A2-S15], this Area measures availability, price and scope, not quality. No independent evaluation of Enhance My Listing output, in any language, was performed.
• A logged-in view of either console. Everything here is from public pages, public APIs, public contracts and public help articles. Seller Central .ae (Ilia's) and noon Seller Lab were not inspected from inside. P1, P2 and P16 all turn on that.
• Legal opinion. This Finding quotes contract language — SPPA §1.3, the AUP clauses, DPP requirements, noon clause 8.3. Whether a given product design breaches any of them is a lawyer's call, and the Agent Policy's technical standards are unread (P13).
• Amazon Accelerate 2026 itself (22–24 Sept 2026). Announcements made there are, by definition, outside this Finding. §1.14 flags the re-check.
• Amazon DSP, AMC and Amazon Marketing Stream as products. Named only as the eligibility gate on Ads Agent; their own onboarding and economics are not assessed.
• Egypt and Namshi as commercial targets. Both appear in the data (1,211 AE-equivalent apps declare EG; noon's API and terms cover EG) and both are outside SCOPE.

---

6. Foreclosure analysis

One line per first-party feature: what a third party should therefore not build, and what stays open. Ordered by how much each actually constrains the SCOPE thesis.

The single genuine constraint on the wedge, stated plainly and without editorialising about the build decision: the "guided process" claim is weakest exactly where it overlaps the Amazon console — Q&A, account health, listing rewrites — because Amazon ships that free, and ships more of it every Accelerate. It is strongest on the pre-account and pre-Amazon half of the spine (explicitly disclaimed by Amazon [A2-S3]), on noon (which ships one button), and in Arabic (which Amazon has deprioritised by policy [A2-S25] and noon has declined by policy [A2-S67]).

And the dated caveat that must travel with all of it: Amazon Accelerate 2026 opens in 16 days, with the Seller Assistant press release's own author on stage [A2-S12]. §1.2, §2.1 and the first two rows of this table have a 16-day shelf life. Re-check after 2026-09-24.