What the guidance establishes

Google says its established SEO practices also apply to AI features in Search. A page needs to be indexed and eligible to appear with a snippet to be eligible as a supporting link. Google does not prescribe a special AI schema or an additional technical requirement for these features.

That is a useful boundary for planning. A proposal to buy a new optimisation service should explain what it changes on your website, which reader need it addresses and how you could inspect the result. A new label is not an explanation of a mechanism.

What it does not establish

Eligibility does not promise inclusion. An observed citation does not establish why the page appeared, and one answer is not a representative view of what all your buyers see.

Treat claims about guaranteed AI citations as a request for evidence. Ask about the questions sampled, the engines and settings used, the comparison period and the changes made at the same time. Without those details, a before-and-after screenshot may be interesting but is not enough to attribute an outcome to a tactic.

Turn the guidance into a page review

Choose one page that matters to an actual buying decision. Work from the reader’s question rather than a checklist of phrases to insert. Our proposed review asks four questions:

Access: can a search engine reach the page and understand its main text?
Purpose: does the page answer a specific question for a specific audience?
Evidence: can a reader inspect the examples, sources or observations supporting important claims?
Action: does the page help the reader compare, decide or do something useful?

An illustrative B2B example

Imagine a software company with a page about choosing a customer data integration. Its current article lists generic benefits and repeats a few product features. The most useful improvement may be to explain a real implementation choice: how to handle a failed sync, who owns the exception and which records can be recovered.

An improved article could show a documented example, state the conditions under which the process works and explain a limitation. These are proposed editorial improvements, not evidence of a citation advantage. Their immediate value is that the reader has more information for a decision.

The same principle applies to comparisons. If you have not tested the products, publish a sourced feature or requirements comparison with that limitation visible. Do not present it as a hands-on verdict.

Keep visibility and business outcomes separate

Google includes traffic from its AI features within the Web search type in Search Console. That reporting should not be treated as a dedicated count of every AI answer mentioning your brand.

Use a small answer observation log alongside your website reporting. The log records questions, answers, dates and source links. Your website data records visits and the actions visitors take. Each can inform an investigation, but neither automatically explains the other.

For a B2B team, a useful conversion might be a relevant newsletter subscription, a return visit to a comparison or an enquiry with a clear business question. Define those actions before evaluating a change. A larger count of mentions can coexist with very little commercial value.

Your next working session

Bring an SEO lead, a content owner and someone who hears customer questions. Choose five recurring buyer questions and identify the page that should help with each. Agree on one page improvement and record the reason for choosing it.

Repeat the same observations after the change, keep a note of other activity and look for patterns worth investigating. This is a learning process, not a controlled experiment by itself. If a claim cannot be supported, keep it as a hypothesis.

The immediate priority is a useful, accessible publication with a clear audience and evidence worth reading. Our AI visibility baseline guide provides the observation fields for your first pass.

Human review in AI content workflows: a practical playbook.