SEO and AI visibility: what agencies should measure

Clients now ask two questions about visibility. The familiar one is “where do we rank?” The newer one is “does ChatGPT recommend us?” Both are reasonable, but they're about different things, and they need different measurements.

This guide sets out what to measure for each, the signals that sit around both, and how to present them without blurring them together.

Start with what each surface returns

A search results page is a ranked list. For a given keyword, location and device, a client's page appears at a position, or it doesn't appear at all. Position is the core unit, and clicks and impressions follow from it.

An AI answer is generated text. It may name a brand, recommend it, compare it with others or leave it out. It may link to sources, and those sources may or may not belong to the client. The same question can produce a different answer on the next run. There is no position in the search-results sense, so borrowing rank metrics for AI answers leads to confusing reports.

What to measure in search

• Positions for tracked keywords. Choose keywords that reflect what the client sells and where. Track them on a regular schedule and look at direction over several weeks rather than single movements.
• Clicks, impressions and click-through rate from Search Console. These show what searchers actually did, including for queries you don't track.
• Technical health. Indexing, canonical tags, sitemaps, broken links and page speed decide whether good content can rank at all.

What to measure in AI answers

Begin with a fixed set of questions that a client's buyers would plausibly ask an assistant. Collect answers for each question from the engines you care about, more than once, and count results per answer.

• Mention rate: the share of answers that name the client. This is the broadest measure of presence.
• Recommendation rate: the share of answers that actively suggest the client for what was asked, not just mention it in passing.
• Citation coverage: the share of answers that link to at least one of the client's own pages as a source.
• Competitor comparison: the same rates for each competitor, measured across the same answers.

Always report the number of answers behind a rate. Thirty percent of ten answers and thirty percent of three hundred answers are not the same finding.

Signals that sit around both

Two further measurements are often lumped in with AI visibility, but they answer separate questions.

• Crawler access: whether the client's robots.txt and firewall let search engines and AI crawlers fetch pages. It's a precondition, not an outcome. A site can allow every crawler and still never be mentioned.
• Referral traffic: visits that arrive on the client's site from AI assistants. It's measured on the site, usually through analytics or a tracking script, and it undercounts because many people read an answer without clicking.

Keep them separate

It's tempting to combine everything into a single visibility score. Resist it. A blended number hides which part moved, and it invites clients to compare things that aren't comparable.

A practical starting set

What not to promise

No one controls what an AI engine says, just as no one controls Google rankings. Set expectations around improving the inputs, such as content, technical health and presence on the sources assistants cite, and around measuring honestly. Avoid promising a mention, a position or a revenue figure.