How to Check If AI Recommends Your Website (and Who It Recommends Instead)

To check if AI recommends your website: write down the five queries a real buyer would type to find a product like yours, look at Google's AI Overview for each one in a clean browser session and note which sites it cites, ask ChatGPT and Perplexity the same questions twice each, test whether AI crawlers can actually read your site, and then repeat the whole thing weekly, because the answers change constantly. That's the entire method. The rest of this post is how to do each step without fooling yourself, which turns out to be the hard part.

Why I care: I run this exact check on founder websites regularly, with the same probe my product uses. In August I checked nine established companies against their own category queries. Eight were not cited in the AI answer for the thing they sell. None of them knew. The AI answers were happily recommending their competitors, aggregator sites, and in one case a company on a different continent. Being invisible is survivable. Being invisible and unaware is how you lose buyers you never knew existed.

Step 1: List the queries a buyer would actually type

Not your brand name. Searching your own brand and finding yourself proves nothing, every brand wins its own name. The queries that matter are the ones typed by someone who does not know you exist yet: "best [your category] tool", "[your category] for [your customer type]", "[competitor] alternatives". Five is enough to start.

One honesty rule here: pick queries with a buyer behind them. "Best B2B software 2026" is a vanity query, nobody with a credit card out types it. The narrower and more purchase-shaped the query, the more an AI answer for it is worth being in.

Step 2: Check Google's AI Overview, and write down who IS cited

Open an incognito window, search each query, and look at the AI Overview box if one appears. Two things to record: whether your site appears in the cited sources, and, more usefully, who appears instead. That second list is the real finding. It is usually some mix of a competitor who did their homework, an aggregator, and a site you have never heard of.

Incognito matters because AI Overviews personalize. Your logged-in browser has spent months learning that you like your own website, and it will flatter you. Location matters too: if your buyers are in the US and you are not, the answer you see at home is not the answer they see. Check from the market you sell to, or use a tool that does.

Step 3: Ask the chat assistants, twice

Ask ChatGPT and Perplexity your buyer queries in fresh sessions with web access on. Then ask again. The second run is not paranoia, it is the point: SparkToro ran nearly three thousand prompts and found the same question almost never returns the same list of brands twice. If a tool or a consultant hands you a single "your AI ranking" number from one run, they are selling you a snapshot of a slot machine.

What you are looking for across runs is the pattern, not the position: do you ever appear, does a competitor appear consistently, does the assistant describe your category the way you do. A competitor who shows up in four runs out of four has something you can go study.

Step 4: Check whether AI can even read your site

Sometimes the reason AI never recommends you is boring plumbing: your robots.txt blocks the AI crawlers, or your content only exists after JavaScript runs and the crawlers never see it, or your structured data tells machines a different story than your pages tell humans. This is the one step you can fully automate: our free 30-second check reads your site the way the AI crawlers do and tells you what they can and cannot see, no signup. (The long version of how AI reads a site is its own post.)

One distinction worth keeping straight while you do this: a citation is your page linked as a source in an AI answer, and it can send you actual visitors. A mention is your brand named without a link, which comes from the model's memory of the wider internet and sends you nothing measurable. Readable pages earn citations. Reputation earns mentions. You want both, but they are built differently, and measured differently.

Step 5: Do it weekly, and date everything

A single check tells you where you stood for one hour on one Tuesday. Kevin Indig's analysis found 40 to 60 percent of cited domains change month to month. That churn is bad news for anyone selling you a stable ranking, but it is genuinely good news for you: the answer box is not settled. The site that gets cited next month is being decided right now, and it can be yours. It also means the useful record is a dated series: same queries, every week, results written down. Trends are real. Single runs are weather.

What to do with what you find

If you are cited: good, now protect it, because the churn cuts both ways. If you are not, the fix is unglamorous and real: pages that directly answer the buyer queries you listed, structured data that machines can parse, crawler access that is actually open, and references from sites the AI already trusts. No secret trick, and be suspicious of anyone selling one. Also be patient on the clock: industry estimates put a brand-new site's first citation months out, so for a young domain the honest expectation is a slow climb, tracked weekly, not a two-week miracle.

And if running five queries weekly in incognito windows sounds like a chore you will do exactly twice: that is the honest reason we built this into Alice. She watches the queries you choose, records who Google's AI cites week over week, and ties it to your own Search Console and GA4 numbers so you can see whether being cited ever turns into actual buyers. But the manual version above works, costs nothing, and will tell you today whether you have a problem.