What AI visibility means, and how to measure it

How SEO, AEO and GEO fit together

SEO: search engine optimization

Helps search engines crawl, understand and rank useful pages in traditional search results.

AEO: answer engine optimization

Makes the information on a page clear enough for a search system or assistant to use in a direct answer.

GEO: generative engine optimization

Examines how a brand and its content appear inside generated answers, including source citations, textual mentions and framing.

These labels overlap. A technically blocked or unindexed page has a weak foundation for every one of them. Google says its AI search features use the same search fundamentals and do not require special AI markup. Read Google's guidance for AI features and websites.

A citation is not the same as a brand mention

A citation is source evidence: the answer provider returned a URL on the audited hostname or a real subdomain. A brand mention is answer-text evidence: the recorded answer named the brand or domain. Either can occur without the other, so combining them into one yes-or-no result hides useful information.

GetSounding keeps mentions and citations separate. The SoundingScore brings these signals together across brand evidence, while each page audit shows the answers and sources returned for one website.

For example, if publisher.example is first and research.example.com is second, an audit of example.com records a citation at position two. If no returned URL matches example.com or a real subdomain, the answer can still count as a brand mention but not as a citation.

What a Google AI Overview check shows

Google AI Overviews can appear inside Google Search and link to supporting pages. A GetSounding check records whether an overview appeared for the tested query and whether its returned sources included the audited site. The report records the market, language, device and date, and keeps “no overview appeared” separate from a failed or unusable check.

What the report records

• The exact question and whether it was a standard or customer-supplied prompt.
• The service and model label used for the controlled check.
• The recorded answer and ordered source URLs returned with it.
• The matching target URL, citation position and reason for the verdict.
• Brand-mention evidence and sentiment only around a confirmed mention.
• A completion state that separates measured negatives from failed checks.

Each result keeps the question, answer, source links and measurement status together. Read GetSounding's full measurement method or run a free audit to inspect the format.

How GetSounding matches citations

This example uses example.com to show the counting rule. The service returns three sources in this order:

1. https://publisher.example/research
2. https://www.example.com/guide
3. https://example.com.attacker.invalid/page

The result is counted as cited, at position 2, matched on https://www.example.com/guide, on a hostname match. Source 2 is a real subdomain of the audited hostname, so it creates the citation and sets the position. Source 3 merely contains the target text: its hostname belongs to attacker.invalid, so it does not match.

If no returned source uses the exact target hostname or a real subdomain, the result has no citation or position. Only the returned URLs determine the citation.

Crawler access and visibility

Crawler rules can affect whether a provider can retrieve or index a page. Source selection still depends on the provider's retrieval system. Search access, user-requested retrieval and model training can use different controls.

Check the provider's current documentation before changing a robots.txt file: the OpenAI crawler documentation, Anthropic's crawler guidance, and Perplexity's crawler documentation.

What optimization can and cannot do

Clear page structure, crawlable text, accurate structured data, useful original evidence and relevant internal links make a page easier to discover and interpret. None of them guarantees a ranking, citation or Google AI Overview appearance. AI answers can change between runs, and a controlled question set is not search-volume data.