The lexicon.
Every term, defined once.

The field has no consensus meaning for this term; this is the definition Rank Sniper states and scores. The SEO pillar (pillar layer v1.0) is 9 checks — Title length and uniqueness, sitemap.xml validity, Meta description, Product type set, Canonical tags, H1 structure, Tag richness, Alt-text coverage and URL structure — with weights summing to 100.

A robots.txt that fully disallows Googlebot and Bingbot costs this pillar 6 points per crawler, capped at 30; search-crawler blocks never touch the composite. Backlinks, domain authority and link graphs are out of scope by choice: Rank Sniper does not fetch, estimate or score them.

The field has no consensus meaning for this term; this is the definition Rank Sniper states and scores. The term has no single canonical origin; it is used loosely across the industry. The AEO pillar (pillar layer v1.0) is 5 checks — Description depth, FAQPage / HowTo schema, List and table density, Answer-first structure and Question-shaped headings — with weights summing to 86.

The FAQPage / HowTo check is scored as evidence that answer content is structured as questions and answers — not as a promise of a rich result. Google removed the How-to rich result in 2023, and FAQ rich results no longer appear in Google Search from 7 May 2026.

The academic term comes from “GEO: Generative Engine Optimization” (Aggarwal et al., accepted to KDD 2024). The field has no consensus meaning for this term; this is the definition Rank Sniper states and scores. The GEO pillar (pillar layer v1.0) is 5 checks — llms.txt presence and quality, Vendor and provenance attribution, Quotable claim density, Freshness (updated_at) and Named entities — with weights summing to 82.

Each of the 10 AI agents the scan checks that robots.txt fully disallows subtracts 6 points from this pillar and from the composite, capped at 30 — a capped penalty in the score. Rank Sniper does not measure whether any engine actually cites a store, and does not claim to know how any engine chooses its sources.

The field has no consensus meaning for this term; this is the definition Rank Sniper states and scores. AIO here means machine readability — whether the record parses and can be acted on. It is not the umbrella definition that treats AIO as all four disciplines at once, and it is not the off-site trust definition.

The AIO pillar (pillar layer v1.0) is 10 checks — Structured attributes, SKU coverage, Variant-level offers, Price with priceCurrency, Availability in schema, Shipping weight, Imagery present, Real variant options, Schema conflict detection and Entity consistency — with weights summing to 108. 5 of them are also scored on their own as the transactability sub-score.

Vendors document their crawlers by purpose, and one operator often runs several. OpenAI lists four tokens and says each setting is independent of the others; Google separates its common crawlers, which honour robots.txt, from user-triggered fetchers, which generally ignore it.

Rank Sniper’s own crawler is RankSniperBot. It reads robots.txt before anything else, stops if it is disallowed, and fetches a handful of public pages per scan rather than crawling the store.

robots.txt matches a crawler to its rules by that token, so spelling matters, and a vendor’s documented token is the one to use. A user-agent string can be spoofed — Common Crawl notes crawlers falsely identifying themselves as CCBot — which is why several vendors, OpenAI among them, publish the IP ranges their crawlers use.

Rank Sniper’s crawler sends RankSniperBot/1.0 (+https://ranksniperhq.com/bot) on every request, with the address of its identity page, so a server log says who it was and how to stop it.

Under RFC 9309 the most specific (longest) matching rule wins, and where an allow and a disallow are equivalent the allow should be used; crawlers should not rely on a cached copy for more than 24 hours. It is not access control: the RFC says it is no substitute for security measures, and Google says it is not a mechanism for keeping a page out of Google — noindex or password protection is.

Shopify generates a default robots.txt for every store; a theme changes it by adding a robots.txt.liquid template. Rank Sniper reads it first, parses it by group, and records which of the 10 named AI agents and Googlebot and Bingbot are fully disallowed.

Shopify’s sitemap links to all products (with the primary product image), pages, collections and blog posts, in child sitemaps by type; it updates as content is added, international domains get their own, and a password-protected store’s sitemap cannot be read by crawlers.

Rank Sniper’s SEO pillar scores sitemap.xml validity (weight 14): a parseable urlset or sitemap index with at least one URL passes; a file that is present but unparseable earns partial credit; a 404 is scored as absent; a fetch that fails outright is unscored. Shopify serves the file, so the fix is guided.

Where several URLs serve the same product and nothing settles which is canonical, the index picks one itself — Google says it will choose the version it judges best. A canonical link states the preference; Google treats it as a strong signal rather than a command.

Rank Sniper’s SEO pillar checks each sampled product page for a canonical link in absolute form (weight 10). Canonical tags are written by the theme template, which the app does not edit, so the fix is guided: the exact tag and the file it belongs in.

Permission and delivery are separate. robots.txt states policy; a firewall, CDN rule or bot filter decides delivery, and can refuse an agent that robots.txt allows. Rank Sniper reads policy today. Testing what each declared AI agent is actually served is an experimental registry check, not scored.

Delivery failures have documented effects: Google does not use content from a URL that answers with a 4xx status, and 429 or 5xx responses make its crawlers slow down. Rank Sniper’s own crawler records which path served each scan; if a declared request is hard-refused it retries that one request with a standard browser profile — never for a path robots.txt disallows.

It is not the same as crawlability. Google states robots.txt is not a mechanism for keeping a page out of Google, and that noindex or password protection is.

Rank Sniper’s registry defines an indexability check — no noindex on live product URLs, and /products/ not disallowed for Googlebot. It is experimental and does not affect any score today.

On Shopify the sitemap is the dependable route, because it is generated automatically and lists every product. Without a valid sitemap an index has to reach the catalogue by following links, and products deep in it are the hardest to reach.

Some engines accept direct notification of changed URLs. IndexNow shares one submission across participating engines; its site lists Microsoft Bing, Naver, Seznam.cz, Yandex and Yep (its FAQ adds Amazon), and Google is not listed.

Google says no special schema.org markup is needed to appear in its generative AI features in Search, and that structured data remains useful for rich-result eligibility. Rank Sniper does not claim otherwise. It scores structured data because it is the form in which a parser can read a price or an availability without guessing.

The scan fetches up to three product pages as raw HTML with no JavaScript executed, so markup a script injects later is not counted. The composite’s structured-data sub-score is out of 100; the AIO pillar adds price with priceCurrency and a check for conflicting prices or availability on one page.

Rank Sniper parses every application/ld+json block as JSON and walks it, @graph wrappers and array @type values included; a malformed block falls back to pattern matching rather than being ignored. It does not read microdata or RDFa.

Nodes can reference one another by @id, which is how a Product, its Organization and its page connect into one entity graph. Checking that graph is an experimental registry check, not scored.

JSON-LD is a W3C Recommendation (JSON-LD 1.1), and Google recommends it for structured data where a site’s setup allows, because the markup sits apart from the visible text and is easier to maintain at scale.

Each type has its own definition. DefinedTerm, for example, is “a word, name, acronym, phrase, etc. with a formal definition”, collected in a DefinedTermSet; this lexicon publishes each entry as a DefinedTerm whose description is the definition shown on the page.

Search engines document which types and properties they use and for what. Google’s merchant listing documentation, for example, requires name, image and offers on a Product.

For merchant listings Google requires name, image and offers on the Product, and a price above zero with an ISO 4217 priceCurrency on the Offer; availability, itemCondition, url, shippingDetails and hasMerchantReturnPolicy are recommended.

Rank Sniper’s composite awards 40 of its 100 structured-data points for Product JSON-LD on a sampled page, 20 for offers, 20 for availability and 20 for shipping details or a return policy. It recognises Product, ProductGroup, ProductModel, IndividualProduct and SomeProducts as product markup. GTIN and MPN are read but not scored in rubric v1.0.

Google’s merchant listing documentation requires price and priceCurrency on the Offer. A price with no currency is ambiguous, and an agent that cannot resolve the currency cannot quote it.

Rank Sniper’s AIO pillar checks that the Product node carries offers (weight 12), that price and priceCurrency appear together (12) and that availability is stated (12). Two different prices or availabilities on one page are scored as a conflict (8), because a parser that finds both cannot pick either.

Google documents it for product variants: name is required; hasVariant holds nested Product nodes; variesBy names the attributes that differ; productGroupID is the parent SKU.

It is the shape Rank Sniper most often finds on Shopify stores with size or colour variants, and it counts as product markup. An earlier version of the parser treated it as no schema at all; that detection bug is fixed.

Google limited FAQ rich results to well-known, authoritative government and health sites in 2023, then stopped showing them: the feature no longer appears in Google Search from 7 May 2026, and its documentation was removed in June 2026. The type itself remains valid schema.org.

Rank Sniper’s AEO pillar scores the presence of FAQPage, HowTo or QAPage in a sampled page’s JSON-LD (weight 20) as evidence that answer content is structured as questions and answers — never as a promise of any rich result. The fix is assisted: the questions and answers have to be the merchant’s own.

Google Merchant Center lists gtin as strongly recommended, not required: omit it if unsure, and supply an MPN where there is no GTIN. OpenAI’s product feed specification requires a gtin or an mpn. Shopify holds it in the barcode field, whose types include GTIN (8, 12, 13 or 14 digits), UPC, EAN, ISBN, ASIN and custom.

Rank Sniper reads gtin and mpn when they appear in Product JSON-LD but does not score them in rubric v1.0. Checking a valid GTIN, or MPN plus brand, per variant is an experimental registry check that needs a connected scan — Connected scan through the Shopify Admin API: Planned.

The kinds are controlled separately. OpenAI states each of its settings is independent, so blocking GPTBot (training) does not block OAI-SearchBot (ChatGPT search). Control tokens such as Google-Extended and Applebot-Extended are not crawlers at all, but robots.txt names that govern how content fetched by another crawler may be used.

The scan checks exactly 10 AI tokens: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, anthropic-ai, PerplexityBot, Google-Extended, Applebot-Extended, Bytespider and CCBot. Tokens not on that list — Claude-SearchBot, Claude-User and Perplexity-User among them — are not checked. Changing the list changes scores, so it moves with a rubric version.

Documented examples: OpenAI’s GPTBot (“disallowing GPTBot indicates a site’s content should not be used in training”), Anthropic’s ClaudeBot (a blocked site’s future materials “should be excluded” from training datasets), Mistral’s MistralAI-Training and Meta’s meta-externalagent. Amazon says Amazonbot may be used to train AI models. Blocking one is a separate decision from blocking the same company’s search or user-initiated agents.

The scan checks GPTBot and ClaudeBot; MistralAI-Training, meta-externalagent and Amazonbot are not on its list. A block on a checked agent costs 6 points under the capped penalty, whatever the reason for it.

OpenAI says sites that opt out of OAI-SearchBot “will not be shown in ChatGPT search answers, though can still appear as navigational links”. Perplexity says PerplexityBot surfaces and links websites and is not used to crawl content for AI foundation models. Anthropic documents Claude-SearchBot, Amazon Amzn-SearchBot, DuckDuckGo DuckAssistBot and Mistral MistralAI-Index for the same class of job.

The scan checks OAI-SearchBot and PerplexityBot; Claude-SearchBot, Amzn-SearchBot, DuckAssistBot and MistralAI-Index are not on its list.

Vendors differ on whether robots.txt applies. OpenAI says robots.txt rules “may not apply” to ChatGPT-User; Perplexity says Perplexity-User “generally ignores robots.txt rules”; Google says its user-triggered fetchers, Google-Agent among them, generally ignore robots.txt; Meta says meta-externalfetcher “may bypass robots.txt”; Amazon says Amzn-User may not follow all robots.txt directives. Anthropic states that its bots honour robots.txt and names no exception for Claude-User.

The scan checks ChatGPT-User; Claude-User and Perplexity-User are not on its list. A disallow for a checked agent counts in the penalty like any other, even where the vendor says the rule may not stop the fetch.

Google-Extended governs whether content Google crawls may be used to train future Gemini models and for grounding; Google says it does not affect inclusion or ranking in Google Search, and it does not govern AI Overviews or AI Mode — Googlebot’s robots.txt rules are the control there, with nosnippet, data-nosnippet, max-snippet and noindex limiting what is shown. Applebot-Extended opts content out of training Apple’s foundation models; Apple says it does not crawl webpages, and disallowed pages can still appear in search results.

The scan checks Google-Extended and Applebot-Extended. A disallow for a checked token costs 6 points under the capped penalty, the same as a crawler block.

It is why one company runs separate agents: a search-index crawler builds what is retrieved later, and a user-initiated fetcher retrieves a page at answer time. A store that blocks both, or serves its product facts only through JavaScript, gives retrieval nothing to return.

How any engine ranks or selects retrieved documents is not published, and Rank Sniper does not claim to know it.

Google uses the term in its crawler documentation: Google-Extended governs whether content may be used to train Gemini models “and for grounding”.

A system can only ground an answer in facts a page actually states. Rank Sniper’s AEO, GEO and AIO checks score whether a product record states them explicitly — price with currency, availability, material, dimensions, a named vendor.

Vendors describe this for their own products: Perplexity says Perplexity-User may “include a link to the page in its response”; OpenAI says sites opted out of OAI-SearchBot can still appear as navigational links.

Rank Sniper scores inputs that make a record quotable and attributable — answer-first descriptions, numbers with units, a named vendor. It does not track citations, does not know how any engine chooses them, and does not promise them.

The term, like “answer engine optimisation”, has no single canonical origin and is used loosely across the industry. Rank Sniper uses it for systems that lift a quotable passage as the answer.

That is what the AEO pillar scores for: Description depth, FAQPage / HowTo schema, List and table density, Answer-first structure and Question-shaped headings.

The term was framed academically in “GEO: Generative Engine Optimization” (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; arXiv 2311.09735, accepted to KDD 2024), which studied how content can be made more visible in generative engine responses.

Google’s generative features in Search are AI Overviews and AI Mode. Google says there are no additional requirements to appear in them and no special optimisations necessary.

Proposed by Jeremy Howard in September 2024 and revised as a v2 of the proposal in August 2026, the format requires only an H1, with an optional blockquote summary and sections of Markdown links. Google states that Google Search ignores llms.txt files, so they neither help nor harm visibility there. The crawler documentation of OpenAI, Anthropic and Perplexity does not mention consuming them.

Rank Sniper’s GEO pillar scores it (weight 20) as a cheap, explicit statement of a store’s terms, not as a known ranking factor: present earns partial credit, present and pointing at the catalogue earns full credit, and an HTML page served at /llms.txt does not count. Publishing it at the domain root is outside an app’s write scope, so the fix is guided.

AIO here means machine readability — whether the record parses and can be acted on. It is not the umbrella definition that treats AIO as all four disciplines at once, and it is not the off-site trust definition.

It depends on what the server returns before any JavaScript runs: the scan reads raw HTML, as a crawler that does not execute scripts would, so structured data injected only by a script is not seen. Contradictory markup is worse than absent markup — a parser that finds two prices cannot pick one — so a conflict is scored as a failure.

Rank Sniper’s GEO pillar scores named entities (weight 14): whether a product’s title or description names a material, standard or provenance — merino, stoneware, OEKO-TEX, “made in”.

The AIO pillar scores entity consistency (weight 8): whether each product’s vendor agrees with the catalogue’s dominant vendors, because inconsistent names split one brand into several as far as a model is concerned.

Rank Sniper’s GEO pillar scores the share of products updated within the last year (weight 16), reading updated_at, or published_at where that is missing. A timestamp should reflect a real change, so the designed fix refreshes only records where something actually changed.

Some engines accept notification of changed URLs: IndexNow shares one submission across participating engines, and Bing recommends it in its AI Performance announcement.

Shopify’s own statements are dated and have changed. Its Winter ’26 Edition (10 December 2025) listed ChatGPT, Perplexity and Microsoft Copilot as available through Agentic Storefronts. Its explainer of 18 June 2026 describes ChatGPT as live for Shopify merchants selling to US buyers, Copilot as live for Catalog merchants selling to US buyers, and AI Mode in Google Search and the Gemini app as available to select brands selling to US buyers; it does not mention Perplexity.

Google’s checkout on eligible listings in AI Mode and Gemini is an early-access programme for select merchants in the US, Canada and Australia. The current status of OpenAI’s Instant Checkout could not be confirmed on an OpenAI page, so this lexicon makes no claim about it. Rank Sniper scores whether a record is complete enough for these surfaces to use; it does not connect a store to any of them.

Two open protocols are documented. The Agentic Commerce Protocol (ACP), developed by Stripe and OpenAI, is “open source and community-designed under the Apache 2.0 license”; OpenAI’s Agentic Checkout Spec lets merchants run checkout inside ChatGPT while keeping orders and payments on their own stack. The Universal Commerce Protocol (UCP), announced by Google on 11 January 2026 and co-developed with Shopify, Etsy, Wayfair, Target and Walmart, covers discovery through post-purchase support and is also Apache 2.0.

Shopify describes Agentic Storefronts as a sales channel in the Shopify admin that connects a merchant’s products to AI shopping platforms. Rank Sniper implements neither protocol. Several experimental registry checks take their pass conditions from ACP’s feed sections; none is scored today.

Google Merchant Center requires id, title (up to 150 characters), description (up to 5,000), link, image_link, price and availability; it strongly recommends gtin and requires mpn where there is no GTIN. OpenAI’s product feed specification accepts JSONL or Google-compatible TSV/CSV, requires the same core fields plus brand and a gtin or mpn, and uses flags to control search and checkout eligibility.

Per Shopify, its Catalog infers categories, extracts attributes and consolidates variants for Agentic Storefronts. Rank Sniper does not generate or submit a feed today; its automatic fix class is designed to publish through its own metafield namespace, structured data or feed.

Rank Sniper’s composite checks that it is set (weight 12), and so does the SEO pillar (12): a product with no type joins no comparison set. Mapping product_type to a Google product taxonomy path is an experimental registry check, not scored.

The fix is assisted: a type derived from the catalogue itself, staged and written only when the merchant approves it.

Rank Sniper reads it three ways: the composite checks it is set (weight 5); the GEO pillar scores vendor and provenance attribution (16); the AIO pillar scores entity consistency — whether each product’s vendor matches the catalogue’s dominant vendors (8).

Where a vendor is blank only the merchant knows the brand, so the fix is assisted: proposed from the catalogue’s own dominant vendors, applied on approval.

A Shopify product can have up to 2,048 variants (raised from 100 in 2025) and up to three options, such as size, colour and material. In structured data, Google documents ProductGroup with hasVariant for exactly this shape.

Rank Sniper’s checks: real variant options rather than a lone option named “Title” (composite weight 8); a SKU on every variant (12); a non-zero weight on at least one variant (9); and, in the AIO pillar, offers in the product markup (12).

Unlike a GTIN, a SKU means nothing outside the store; its job is to be stable. Rank Sniper’s composite passes a product only when every variant carries a non-empty SKU (weight 12); the same check counts in the AIO pillar (14) and in the transactability sub-score.

Only the merchant can assign one, so the fix is assisted: the field is staged on every variant that lacks it, for the merchant to fill and apply in one pass.

The public scan does not read collections today. The registry defines experimental checks that would, such as breadcrumb markup consistent with collection structure; none is scored.

Rank Sniper’s SEO pillar scores URL structure (weight 8): whether a handle reads as words rather than an imported identifier. Single-word handles pass.

Changing a handle changes a live URL that external links and ads may point at, so the fix is assisted and never applied without the merchant.

Rank Sniper’s automatic fix class is designed to write only to its own metafield namespace, structured data or feed — never to a field the merchant wrote — which is why it is designed to need no approval.

Whether structured attributes held in metafields are actually exposed to storefront surfaces is an experimental registry check that needs a connected scan.

No current Rank Sniper check reads metaobjects, and the public scan does not fetch them.

Shopify’s documented storefront API for a single product is the Ajax Product API (/products/{handle}.js); /products.json is described only by third parties. Rank Sniper’s free scan reads it because it is what a public client can reach, paginating in store order and spreading its sample across the catalogue rather than reading only the first, best-kept page.

Rank Sniper observes fields such as title, handle, body_html, product_type, vendor, tags, variants with sku and grams, options, images and updated_at. Its image objects carry no alt text, so alt coverage is measured from product-page HTML instead. A robots.txt rule disallowing /products.json stops the catalogue read, and the scan with it.

It exposes the robots, group, rule, user_agent and sitemap Liquid objects, and Shopify strongly recommends building on them rather than replacing the defaults, which it updates to keep current.

Unblocking an AI agent on a Shopify store is therefore a theme change. Rank Sniper’s crawler-access fix is guided — the exact rules and where they go — because the app does not edit themes.

Shopify describes theme app extensions as the way merchants add an app’s elements to a theme without editing Liquid templates or code; an app block is positioned in the theme editor like any other block.

Rank Sniper’s automatic structured-data fix is designed to publish Product JSON-LD this way, from facts already in the catalogue, so no theme file and no native product field changes — Automatic fixes — app-block structured-data markup: Planned.

It reaches what a public scan cannot: GTIN or MPN per variant, inventory state, dimensions, metafields. Rank Sniper’s connected scan is designed to read the catalogue through it with the merchant’s permission; 18 of 58 registry checks depend on it.

Nothing reads a store through it today, because the app is not installed on any store — Connected scan through the Shopify Admin API: Planned; Install from the Shopify App Store: In development.

Scopes bound what an app can do, which is why Rank Sniper’s design classes each fix by whether it falls inside the app’s write scope: a change outside it, such as a theme template or a file at the domain root, is guided rather than written by the app.

The app cannot be installed today, so it holds no scopes on any store — Install from the Shopify App Store: In development.