Google Names 6 AI Search Myths to Ignore

What Google confirms, and what it does not

Google has expanded its guidance significantly since the first version published in May 2026. As of August 26, 2026, its position is more precise than most generic GEO checklists: AI Overviews and AI Mode still depend on the foundations of Google Search. Google does not require a secret file, a writing format for language models, or a new schema type.

Google’s official guide confirms five useful points:

1. Generative Search features rely on core Search ranking and quality systems.
2. They retrieve recent pages from the Search index before generating an answer.
3. A page still needs to be crawlable, indexable, and eligible for a snippet.
4. Original, useful content based on real experience is more valuable than another summary of existing material.
5. Performance can now be monitored in a dedicated Search Console report when the property has access.

Google also names tactics that are not required: llms.txt, AI text files, Markdown alternates, artificial content chunking, and manufactured mentions.

This position is about Google Search. It does not describe ChatGPT Shopping, Claude, Perplexity, a browser agent, or Shopify catalog interfaces. Treating one system’s documentation as a universal AI rule is the first mistake to avoid.

AI Overviews and AI Mode are live in France

Google launched AI Overviews and AI Mode in France on July 22, 2026 across mobile, desktop, and the Google app.

The surfaces are not identical:

• an AI Overview may appear above results when Google considers a generated summary useful;
• AI Mode is a conversational Search experience where the user can continue exploring the question.

The launch does not mean that every French query triggers an AI Overview. It also does not create a way for a brand to request or guarantee inclusion.

RAG and query fan-out: multiple searches, multiple sources

Google documents two central mechanisms.

The first is retrieval-augmented generation, or RAG. Search systems retrieve relevant, recent pages from the index. Those pages ground the answer and provide links to supporting sources.

The second is query fan-out. The model generates multiple related searches in parallel. A request such as “a waterproof hiking bag for a 16-inch laptop delivered before Friday” may lead to separate searches for dimensions, waterproofing, delivery, and reviews.

The operational consequence is simple. One answer may draw from a product page, a return policy, Merchant Center, a comparison article, a review page, or another domain. Google does not say that all those sources must be identical. But when they disagree about the brand, product, price, or policy, the system has to resolve conflicting facts.

JavaScript: Google can render it, but complexity remains

The previous version of this article said too broadly that AI crawlers did not execute JavaScript. That was inaccurate for Google.

Google says it can process content in JavaScript when the required resources are not blocked. It also notes that SEO for JavaScript frameworks is more complex than for pages where the essential content is directly available.

For a Shopify store, the robust rule is not “Google cannot see JavaScript.” The better rule is:

• price, availability, variants, and specifications should be reliably rendered and indexable;
• the server response should include the facts needed by systems that do not render the page whenever possible;
• visible values, JSON-LD, Shopify, Merchant Center, and policies should not contradict one another.

Server rendering does not guarantee a citation. It removes an extra dependency and makes the same fact available to more systems.

Schema.org: useful, but not required for generative Search

Google explicitly says structured data is not required for generative Search and that there is no special schema.org markup to add.

That does not make structured data useless. It has specific purposes:

• eligibility for supported rich results;
• describing a Product, Offer, price, availability, or AggregateRating;
• identifying an Organization or OnlineStore;
• publishing return and shipping policies;
• complementing data submitted through Merchant Center.

The distinction matters. Perfect Product markup is not a guaranteed ticket into an AI Overview. It remains a standard way to describe a product and reduce ambiguity on surfaces that use the markup.

llms.txt: no effect on Google Search visibility

Google says Search does not use llms.txt, special Markdown files, or other AI text files to rank or include a site in generative features. Publishing them neither helps nor harms Google Search visibility.

The Lighthouse agentic browsing documentation nevertheless checks llms.txt as an optional convention. It says the file can help some agents find a site’s main structure faster. When the file returns 404, the audit is simply not applicable.

There is no contradiction. Google Search and a browser agent are different surfaces.

No current public documentation from OpenAI, Anthropic, or Perplexity guarantees that a merchant’s llms.txt file will be consumed. A platform publishing its own file is not evidence that it reads files from other sites.

Search Console now measures generative AI features

Google provides a Generative AI Performance report. It can show:

• impressions;
• pages;
• countries;
• devices;
• dates.

This first-party property data is more robust than manually replaying a small prompt set. It still does not measure a recommendation share or prove that a specific site change caused an impression.

Access is rolling out. A property that does not yet see the report should not conclude that its content is absent from generative features.

The new inclusion control is still limited

Search Console also provides a Generative AI in Search control. As of August 26, 2026, Google is testing it with a subset of site owners.

The default setting includes the property’s links and content in the covered generative features. Exclusion prevents the site’s content from appearing as a link or grounding those answers. Google says this choice:

• is not a ranking signal for other parts of Search;
• does not replace Merchant Center or Google Ads settings;
• does not control model training;
• may take several days to take effect.

For an ecommerce brand seeking visibility, the first check is whether the property was excluded by mistake, if the control is available.

Shopify UCP: an agentic layer separate from Google Search

Google names UCP among emerging technologies for agentic experiences. UCP provides a shared language for discovery, catalogs, and other commerce operations. It was co-developed by an industry group that includes Google and Shopify.

On Shopify, the testable surface is Storefront Catalog MCP. Catalog calls use:

https://your-domain.com/api/ucp/mcp

They require a valid agent profile. The documented tools are:

• search_catalog for product discovery;
• lookup_catalog for resolving products or variants by identifier;
• get_product for product detail and variant selection.

Some stores may restrict access. A valid endpoint response proves that a catalog interface was available at test time. It does not prove that a product will be cited, ranked, or purchased.

UCP is not documented as a Google Search ranking factor. It solves a different problem: giving compatible agents a structured way to query a catalog.

ChatGPT Shopping: verified status on August 26, 2026

The previous version incorrectly stated that Instant Checkout had closed in March 2026. OpenAI documentation updated August 21, 2026 says it may still be available for some eligible products and merchants.

OpenAI also says:

• Shopify product data is integrated through Shopify Catalog, with no individual merchant action required for that integration;
• product selection considers the query, context, and metadata from direct or third-party sources;
• product titles and descriptions may be simplified by the model;
• review summaries can come from public websites and are not verified by OpenAI;
• price or shipping changes may take time to appear.

This is not how a Google AI Overview works. The two surfaces must be tested and described separately.

When several sources identify the wrong entity

More sources do not automatically create truth. An answer may retrieve relevant pages and then attach a fact to the wrong company or product.

A recent German judgment documents this risk. In case 26 O 869/26, a Google AI Overview produced associations that were not present in the cited sources when considered separately.

For a Shopify brand, the sensitive fields are concrete: public name and legal name, primary domain, official profiles, product brand, GTIN, MPN, SKU, canonical URL, and review source.

The complete method and its limits are in When Google AI Overview Confuses Two Companies.

Priorities for a Shopify store

Priority 1: verify access and identity

• Confirm that important pages are indexable and snippet-eligible.
• Check the site name, canonical domain, and Organization or OnlineStore entity.
• Review the new Search Console inclusion control when available.

Priority 2: align product facts

• Compare price, stock, variants, brand, and identifiers across Shopify, visible content, JSON-LD, and Merchant Center.
• Check shipping, returns, and reviews separately.
• Resolve contradictions before producing more content.

Priority 3: measure each surface with its own evidence

• Use the Generative AI Performance report for Google.
• Test Storefront Catalog MCP with a valid agent profile for Shopify UCP.
• Inspect ChatGPT Shopping without using it as evidence for AI Overviews.
• Record date, country, language, query, product, and observed sources.

Priority 4: publish content that is difficult to replace

Create real tests, original data with a method, expert answers, and useful images. Google advises against creating pages only to cover every possible query fan-out variation.

The Verity Score Engine Coverage panel

AI systems do not consume the same surfaces. The Verity Score Engine Coverage panel separates signals by target engine and shows the supporting doctrine when a source is available.

The panel is designed to prevent a misleading conclusion such as “this signal matters to every AI engine.” It describes documented applicability or evidence status. It does not convert a technical signal into a visibility promise.

The Verity Score Shopify app compares facts exposed in pages, structured data, reviews, and available commerce surfaces. Merchants review proposed changes before publication.

Limits of this guide

This guide describes public documentation as of August 26, 2026. The Search Console report and inclusion control are still rolling out. UCP is changing quickly. Platforms may change their interfaces, eligibility rules, or sources.

None of these actions guarantees an AI Overview placement, a ChatGPT citation, or a product recommendation. They make facts easier to access, more consistent, and easier to attach to the right entity. That is a verifiable prerequisite, not a result promise.