The useful thing about AI search is that the output is public. You do not have to take a report on trust, because you can go and look at the same answers the report is describing.
Four checks, about twenty minutes, no special access required.
Check 1: run their prompts yourself
Take the prompt list from your report, open a logged-out or temporary chat, and ask five of them.
Logged out matters more than anything else here. In your normal account the assistant has seen you discuss your own business and will bring it up. That is memory, not visibility, and it is the single most common reason people believe things are going well when they are not.
Record two columns: were you named, and did your domain appear in the citations. Compare against what the report says. They should broadly agree, allowing for the fact that these systems vary day to day.
If the report shows movement you cannot reproduce at all, ask how the measurement is done: which engines, how often, logged in or out, and how a mention is counted.
Check 2: look for the two-column split
Ask whether the report separates being cited as a source from being named in the answer.
These are different outcomes with different causes. Being cited means your page was used as reading material. Being named means your business was recommended. Across the 608 answers we measured in one category, the set of brands supplying the reading and the set getting the recommendation overlap far less than anyone expects.
If your report gives one blended number, it cannot tell you which of two very different situations you are in, and they need close to opposite work:
- Cited but not named: your content is doing its job. Your presence in other people's content is not. The fix is almost entirely off your website.
- Neither: you are not being retrieved. The fix starts on your own pages and is usually the faster of the two.
This is the single most informative question you can ask about a report.
Check 3: check the claiming actually happened
Pick three citation sources your category runs on and look at your own listing on each. Is it there, complete, and consistent with the others?
This is the least glamorous workstream and usually the highest value. It is also completely verifiable in ten minutes, and it is the part most likely to be described in a report and not actually done.
What to look for: the business name spelled identically everywhere, the same address format, the same phone, current opening hours, a description that matches what you actually sell, and categories that are specific rather than generic.
Check 4: search for yourself somewhere that is not your site
Look for mentions of your business in the last three months on sites you do not own. Round-ups, comparisons, community threads, industry lists.
“Customers were asking ChatGPT and Gemini for product recommendations in our exact category, and we weren't even in the conversation. Within three weeks of launching Omni Eclipse's AEO strategy, we were the most-cited brand in the space - ahead of long-established brands with far more online presence.
The sources that decide who gets named are mostly not yours. If nothing has appeared anywhere external in six months and your named figure is flat, that is very likely the reason, and the answer is not more articles on your own site.
Two metrics that mean nothing on their own
Traffic. AI search often produces no click at all. Someone hears your name, remembers it, and arrives directly a week later. Traffic can be flat while the work is going well, and it can rise for entirely unrelated reasons. It is not a proxy here.
Articles published. Volume is an input. Twenty pieces that get no lift is worse than four that do, because it also costs you the time to review them.
The honest timeline
Weeks 1 to 6: citations should start moving if the claiming work was real. The models re-read those sources often. If nothing at all has changed by week eight, ask specifically what was claimed and check three of them yourself.
Months 2 to 4: content-led visibility. Slower, because a page has to be published, indexed and then retrieved.
Months 3 to 6: being named. This depends on other people's content and it is the slowest of the three. Judging the engagement before month three is not fair. Judging it at month six is.
What the upper end looks like
Worth knowing, because it calibrates what "slow" actually means.
Fur Magic's customers were asking ChatGPT and Gemini for recommendations in their exact category, and Fur Magic was not in the conversation. Three weeks later they were the most-cited brand in the space, ahead of long-established competitors with far more online presence.
Annie Belle Boutique went from nobody finding them on AI search to being recommended in almost half of relevant queries, in six days.
Both are published with the starting number, the end number and the period, so you can apply check one to us: Fur Magic and Annie Belle Boutique.
When to have the conversation
Citations up, named flat, at month six. The work is happening and the strategy is wrong. This is a real conversation and a fixable situation, and it almost always means the third-party workstream has been thin. Ask what external placements have been made, and what the next three are.
Neither has moved and you cannot verify the claiming. Different conversation. Ask for the list of sources claimed with dates, and spot-check three.
Everything moving but you cannot see it between reports. Ask for a login rather than a deck. A monthly report is a curated view assembled by the person whose work it describes.
The four checks, run on a real engagement
What running all four in one sitting actually looks like.
Check one. Five of the twenty tracked questions, logged out, across three engines. Named in two, cited in four. The report claims named in three. That is within normal variance for a single run and not a discrepancy worth raising.
Check two. The report has separate columns for cited and named, with month-on-month deltas on each. It passes. If it had shown one blended score, everything else would be unverifiable.
Check three. Three listings opened side by side. Two are correct and current. The third still shows a phone number changed eighteen months ago, and it is on a source the report claims was corrected in month one. That is the finding worth raising, and it took four minutes.
Check four. Searching for the business name outside its own domain turns up one industry round-up from month three and nothing since. The named figure has been flat for two months. Those two facts are the same fact.
The conversation that follows. Not "is this working". Specifically: the third listing, and what the next three external placements are. A good agency answers both in the same email.
Get a second opinion on your own numbers
If you have run the four checks and something does not add up, we will run your prompt list independently across all five engines and show you what we see, with the method written down so you can reproduce it.
Get an independent read
A free AI visibility audit on your own buying questions, across all five answer engines.
Book your audit
Ashur Homa
Built and scaled a digital brand to $100M+ in sales with zero ad spend. Has helped businesses generate millions through AI go-to-market strategy. Leads growth at Omni Eclipse.
Connect on LinkedIn
