How to check whether AI recommends your software
Godfrey, lessClick
You can check whether AI assistants recommend your software in about 30 minutes. Write ten questions in your buyers’ words, ask each one three times in ChatGPT, Gemini and Google’s AI Overviews, and log who gets named, in what order, and which sources the answer cites. The log is your baseline.
Which questions should you ask?
Ask the questions a buyer types before they know your name. “Best contract software for a 40-person company” is a buyer question. “Is [your product] any good” is not, because the buyer already found you.
A useful set of ten mixes four kinds of question:
- Category questions. “What is the best [category] for [type of company]?”
- Problem questions. “How do I [job your product does] without [common pain]?”
- Comparison questions. “Alternatives to [the market leader]” and “[competitor] vs [competitor]”.
- Constraint questions. Price, team size, region, integrations: “[category] that works with [tool] under [budget]”.
Take the wording from sales calls, support tickets and the search terms in your own analytics. Buyers rarely use the words on your homepage.
Which assistants should you test?
Test the assistants your buyers use. For most B2B software companies that means ChatGPT, Gemini and Google’s AI Overviews, the summary that appears above Google’s results. Perplexity and Claude are worth adding if your buyers are technical.
Use each assistant the way a buyer would: the free or standard version, logged out or in a fresh session, with no custom instructions. An assistant that knows you work at the company can give a different answer from the one a stranger gets.
Why ask each question three times?
Assistants do not give the same answer every time. The same question, asked twice in a row, can return a different list of names. One run tells you what happened once. Three runs show you whether a name appears reliably, sometimes or never.
Open a new chat for each run, so the earlier answer does not shape the next one.
What should you write down?
Log every run in a spreadsheet with one row per answer. These columns are enough:
| Column | What to record |
|---|---|
| Date | The day you ran it |
| Assistant | ChatGPT, Gemini, AI Overviews |
| Question | The exact wording |
| Run | 1, 2 or 3 |
| You named | Yes or no |
| Position | Where your name appears in the list, if at all |
| Companies named | Every company in the answer, in order |
| Sources cited | The pages the answer links to or names |
The last column matters most. The sources tell you which pages the assistant leans on: your own pages, a competitor’s comparison page, a review site, a directory or an article. Those are the pages that decide the answer.
How do you read the results?
Look for three patterns once the log is full.
Questions where you never appear. These are the gaps. Note who is named instead and which sources the answer cites.
Questions where you appear sometimes. These are the closest wins. Something about your pages or your presence on other sites is nearly enough.
Sources that keep coming up. A comparison article or directory that appears across several questions is worth more attention than any single page on your own site.
What should you do after the test?
Run the same questions again in a month, with the same wording, in the same assistants. Comparing month to month only works if the questions stay fixed, so resist the urge to edit the list after the first run.
If the first run shows you named nowhere, that is a normal starting point and a useful baseline. lessClick’s free snapshot runs five of these questions for you and sends back one page: who AI names, and where you stand.