AI Visibility: Measuring Whether AI Search Knows Your Store

The measurement problem

Traditional SEO has a number. You are third for a keyword, or you are not, and a tool tells you which. AI search has no equivalent. There is no position, no dashboard, and the same question asked twice can return different sources.

That is why most stores have no idea whether assistants mention them. Not because they are careless, but because the obvious way to find out does not exist. What replaces it is sampling: ask the questions your buyers ask, repeatedly, and record what comes back.

It sounds crude. It is also the only honest method, and once you have a few weeks of it, the picture is clearer than a rank tracker ever gave you, because you are measuring what was actually said about you rather than where a link sat.

What AI visibility actually means

Being cited

Your page is used as a source and linked. This is the strongest outcome and the easiest to verify, since engines like Perplexity show their citations openly.

Being mentioned

Your brand is named in the answer without a link. Weaker on paper, but frequently more valuable, because a recommendation inside an answer carries more weight than a link beside it.

Being described accurately

The engine knows what you sell, who you serve, and what makes you different. Many stores are known but described wrongly, which is a distinct problem with a distinct fix.

Being absent

The engine answers the question using competitors, publishers or marketplaces and never reaches you. This is the default state for most stores, and it is a content problem rather than a technical one.

Building a question set

Everything downstream depends on asking the right questions, so this is worth doing carefully rather than quickly.

1. Start with your buyers. What do they ask before purchasing? Support tickets and pre-sale emails are the best source, because they are real language rather than keyword-tool language.
2. Include category questions. "What is the best X for Y" and "which X should I buy for Z". These decide purchases.
3. Include problem questions. People often describe a symptom rather than a product. "Why does my X keep doing Y" reaches buyers earlier than any product query.
4. Include comparison questions. "X vs Y" and "is X worth it". High intent, and assistants answer them readily.
5. Include your brand. "Is Brand legitimate", "what does Brand sell", "Brand returns policy". You need to know what is being said.

Twenty to thirty questions is enough. Fewer and the sample is noise. More and you will stop running it, which is worse than a smaller set run consistently.

Running the check properly

Always start a fresh session

An assistant that already discussed your brand will mention it again. Every check must start clean or the result is worthless.

Ask the question as a buyer would

Not "best organic coffee beans UK site:mystore.com" but the sentence a person would type. Keyword syntax produces answers no real user will ever see.

Record more than yes or no

Note whether you were cited, mentioned or absent, which competitors appeared, and which sources were used. The competitor column is where the useful information is.

Sample repeatedly

Answers vary between sessions. One check tells you almost nothing; five spread over a fortnight tells you whether you are genuinely in the mix or got lucky once.

Keep the record

A spreadsheet is fine. The value is in the trend, and you cannot see a trend from memory.

Reading the results

The last row is the most common and the most fixable. Open the page that was cited, find the exact passage, and compare it with the equivalent section on yours. The difference is usually structural rather than a matter of authority.

What moves the number

Answering questions directly

The single largest factor. A section that opens with the answer beats one that builds toward it, every time, on every engine.

Specific and checkable claims

Numbers, materials, timeframes, conditions. Assistants relay concrete detail and skip over generalities, because generalities are not worth relaying.

Topic depth rather than product pages

Stores that only publish product pages are invisible for informational questions, which is most of them. Depth on the subject is what gets you into the pool.

Consistency across your site

If three pages give three different answers to the same question, none of them is trusted. Contradictions are more damaging here than thin content.

Being mentioned elsewhere

Assistants draw on the whole web, not just your site. Reviews, forums, supplier pages and press all feed the picture. You cannot control these, but you can be worth mentioning.

Fixing an inaccurate description

This deserves separate treatment because the instinct is usually wrong. When an assistant describes your store incorrectly, the cause is almost never malice or a bug. It is that the correct information is not stated plainly anywhere it can find.

Say you stopped selling a category two years ago and assistants still list it. Somewhere there is an old page, a directory listing or a review that says otherwise, and nothing on your site contradicts it clearly. The fix is a page that states what you sell now, in plain sentences, on your own domain.

The same applies to shipping, returns, and whether you are a real business. If your policies live only inside a checkout flow, no engine can read them. Put them on a page, write them as sentences rather than bullets of legal text, and the description improves within weeks.

How often to check

Monthly is enough for most stores, with a fuller review each quarter. Weekly checking produces noise that looks like movement and encourages changes that were never justified.

The exception is after publishing significant content or correcting a factual error, where a check two to three weeks later tells you whether the change was picked up.

What Rankmore does here

Rankmore checks Perplexity weekly against your campaign keywords and records whether your store appears, so you get a trend without running the sampling by hand. Coverage of other engines is rolling out.

That does not replace your own question set, and it is not meant to. Automated checks track the keywords you are actively working on; your manual set covers the awkward, specific, human questions that never look like keywords and are often where the buying decisions happen.

Turning a check into a work list

Sampling only pays off if the results become tasks. After each monthly round, the record should produce three short lists.

Questions you lost narrowly

You appeared sometimes, or a competitor was cited from a page barely better than yours. These are the cheapest wins, because the page already exists and already ranks well enough to be considered. Tighten the passage and re-check next month.

Questions you have no page for

You were absent every time and nothing on your site addresses the question. These need writing, and they should be prioritised by how close the question sits to a purchase rather than by how often it is asked.

Facts that came back wrong

Anything the engine stated about your business that is not true. Treat these as urgent regardless of volume, because a wrong answer about shipping or legitimacy costs more than a missing answer ever does.

Three lists, worked through in that order, is the whole programme. It is unglamorous and it compounds, which is a fair description of most things that work in search.

What good looks like after six months

Realistic expectations help here, because the early weeks look like nothing is happening.

By month two you should see occasional appearances on the narrow questions where you genuinely know more than a generalist publisher does. By month four, appearances on those become common rather than occasional, and brand descriptions read correctly. By month six, you should be appearing on at least some category questions, which are the hardest and the most valuable.

What you should not expect is to displace large publishers on broad questions like "best coffee grinder". That is a fight worth losing gracefully. The specific questions underneath it, where you have real handling experience, are both winnable and closer to a sale.

Common mistakes

• Checking once and concluding. Variance between sessions is large. Sample repeatedly or do not bother.
• Testing in a polluted session. If your brand came up earlier in the conversation, the result is meaningless.
• Only tracking your own name. Brand queries are the easy ones. Category questions are where new customers are.
• Ignoring who won instead. The cited competitor is the most useful output of the whole exercise and most people never open it.
• Expecting a single score. There is no rank here. Direction over months is the honest measure.

Final thoughts

AI visibility is measurable, just not the way rankings were. Ask real buyer questions in clean sessions, record what comes back including who won instead, and repeat monthly.

Most stores that do this discover the same two things: they are absent from every category question, and the pages beating them are not better resourced, just better structured. Both are fixable with editing rather than budget.