Business visibility in Google AI Mode: what works and what is just a myth

“AI SEO” is often presented as a new channel with its own admission ticket: an llms.txt file, special schema and pages rewritten into small answer-shaped blocks. That is not how Google documents its generative Search features. To be eligible as a supporting link in AI Overviews or AI Mode, a page must first be indexed, eligible to appear with a snippet and compliant with the usual Search requirements. Google does not require an AI text file, AI-specific markup or a secret content format.

Something has changed, nevertheless. A generative result can break a complex question into related searches, combine sources and place links in a different context from a conventional results page. Discovery and measurement are evolving. The foundations remain familiar: accessible pages, clear information, credible evidence and a useful experience after the click.

For a small business, the sensible response is not to buy a bundle of speculative GEO tactics. It is to establish whether the website is eligible and worth using as a source, then measure what changes.

Start with the claims Google actually makes

Google describes AI Overviews and AI Mode as parts of Search. They use the Search index and core ranking systems, although different models and link-selection techniques may be involved. AI Mode is particularly suited to exploratory, comparative and multi-part questions, where its systems can run several related searches before composing a response.

That supports four careful conclusions:

• there is no guaranteed submission process for inclusion in AI Mode;
• technical compliance creates eligibility, not a promise of visibility;
• a conventional ranking and an appearance as a supporting link are not identical user experiences;
• a handful of manually repeated prompts cannot prove overall performance because responses and citations can vary.

Nor is there evidence that adding FAQ markup, increasing the number of headings or publishing a fixed volume of articles guarantees selection. Those choices may improve a page when they serve its reader, but the format alone is not the source value.

Myth-versus-evidence decision table

Use this table to separate documented work from experiments that should not be sold as Google requirements.

This does not prove that every experiment is useless outside Google. It stops an unverified tactic from being treated as the price of admission to Google Search.

A six-layer generative-search visibility audit

Rather than auditing an imagined model preference, examine the chain from technical eligibility to business measurement. Each layer answers a different question, and later work cannot reliably compensate for a failure in the first one.

1. Can the priority page be found and indexed?

Check its HTTP status, robots.txt, noindex, canonical URL, rendered content and internal links. A priority URL should be reachable through a normal link, and its main answer should be available as text rather than hidden behind a widget interaction.

An XML sitemap does not force indexation, but it can provide a clean inventory of canonical URLs you want crawled. The free website audit and sitemap inspector can support a first pass. Google’s URL Inspection and index reporting remain the appropriate places to check its actual decision.

2. Does the page have a clear job?

A visitor should quickly recognise whether a page is a service, guide, comparison, product or piece of documentation. The title, primary heading, introduction and sections need to resolve the same task. A broad page that tries to cover every adjacent phrase often becomes a weak answer to all of them.

The on-page SEO inspector can identify basic structural signals such as titles and headings. It cannot decide whether the page resolves a customer’s decision. That requires comparing the content with customer questions, search results and the intended role of neighbouring pages.

3. Does the page offer non-commodity evidence?

A source becomes more useful when it provides verifiable information that another generic summary cannot easily replace. This does not always require an expensive research programme. Strong assets can include:

• a precise compatibility or cost table;
• a stated comparison method and its limitations;
• a dated record of changes to a standard;
• a calculator or working tool;
• a diagnostic sequence that maps a symptom to likely causes;
• original product documentation, imagery or attributed expert answers.

Choose the evidence before choosing the format. Adding citations without understanding them—or manufacturing “proprietary data”—damages the trust the exercise is meant to build.

4. Are the business and offer consistent?

The company name, location, service scope, contact details, prices, availability and policies should not contradict one another across the website, Business Profile, product feeds and structured data. For a local service business, an accurate service page with a defined area, process, authorship and contact route may be more useful than another series of articles about AI.

Consistency is not a guarantee that a business will be recommended. It removes ambiguity that also obstructs a human customer making the same decision.

5. Does structured data describe visible truth?

Structured data can help Google understand supported information types and can make a page eligible for particular search treatments. It is not a private channel for claims a visitor cannot see. Mark up a product as a product, an organisation as an organisation, and an article as an article only when the required properties are accurate and represented on the page.

The schema markup generator can help form a valid JSON-LD block. Then compare the result with the visible page, the documentation for that specific type and official validation tools. Do not invent ratings, authors or services to fill properties.

6. Can exposure and business effect be measured?

Google says its generative AI performance report completed worldwide rollout on 31 August 2026. It includes AI Overviews and AI Mode impressions, with breakdowns by page, country, device and date. That provides a much stronger baseline than manually searching a few prompts.

The report does not, by itself, show whether an impression produced a qualified enquiry, booking or sale. Connect it with:

• release annotations for meaningful site changes;
• conventional Search performance for the same page groups;
• landing-page sessions and conversions;
• CRM or revenue data where its quality supports comparison;
• campaign, seasonality and offer changes that might explain movement.

Search Console also applies reporting limits and data thresholds. A missing row should not automatically be interpreted as proof of zero exposure. Compare meaningful periods and groups of related pages rather than reacting to one day.

A 30-day work plan, not a promise of results in 30 days

The schedule below orders the work. It does not impose a deadline on crawling, selection or visibility.

1. Record the baseline. Save generative and conventional visibility, indexation and conversion data for priority pages.
2. Remove eligibility blockers. Correct broken statuses, accidental noindex, conflicting canonicals, missing links and essential content absent from rendered HTML.
3. Assign pages to customer decisions. Consolidate duplicates, identify missing resources and select one priority cluster.
4. Create source value. Build a table, tool, method, current comparison or piece of documentation that answers a real question.
5. Align business information and structured data. Reconcile visible content, profiles, feeds and markup.
6. Measure after release. Annotate the change and allow for crawl delay, reporting limits and seasonality before drawing a conclusion.

Technical and editorial work need to support one another. A well-written page cannot override an indexation block, while excellent code cannot supply the useful information a searcher needs.

When limiting generative use is the right decision

Maximum visibility is not every organisation’s only objective. Google supports nosnippet, data-nosnippet, max-snippet and noindex controls that also affect presentation in generative Search features. Restricting available snippets can therefore restrict how a page participates in those experiences.

This is a product and policy decision, not a cosmetic SEO setting. A publisher, retailer and professional service may balance discovery and content control differently. Before applying a site-wide directive, identify the affected sections, the risk being addressed and the measurement that will reveal the result. Crawlers also need access to read a directive; blocking the crawl can prevent that instruction from being seen.

The most useful next step

Do not begin by buying a package of “100 AI citations”. Choose five pages closest to an enquiry or sale and assess all six layers: eligibility, purpose, source value, business consistency, structured data and measurement. The output should be a list of specific blockers with owners and priorities, not an abstract GEO score.

If the constraints sit in architecture, rendering, structured data or analytics, I can help turn that audit into working changes through my design and development services. You can also describe the page and business outcome to scope the implementation without promises of rankings or citation.

Sources

• Google Search Central: optimising for generative AI features
• Google Search Central: AI features and your website
• Google Search Central Blog: generative AI performance reports in Search Console
• Search Console Help: generative AI performance report for Search
• Google Search Central: creating helpful, reliable, people-first content
• Google Search Central: general structured data guidelines
• Google Search Central: robots meta and data-nosnippet specifications