Find the buyer questions AI recommends you for, and the ones it gives to competitors.
Ask a fixed set of real buyer questions on every engine, repeatedly, and record for each one whether you were named and who was named instead. A question recommends you consistently only when the interval on its mention rate stays above half. Proofsource does this on ChatGPT, Perplexity, Claude and Google AI Overviews.
How can I check which buyer questions consistently get my company recommended by AI assistants versus competitors?
Treat your buyer questions as a fixed benchmark, not a one-off check. The same question asked twice can return two different shortlists, and the four engines disagree with each other often. So the method has three parts: a stable set of questions, the same questions on every engine, and enough answers per question to tell a pattern from luck.
"Consistently" needs a number, or every lucky answer reads as a win. We use the 95% Wilson interval on a question's mention rate:
- Won when the bottom of the interval is above 50%: the engines recommend you more often than not, and the data says so.
- Lost when the top of the interval is below 50%: rivals take the question and more answers will not rescue it.
- Contested in between: act on it only after more answers.
How do you find those questions, step by step?
- Collect the questions buyers really ask. Pull them from sales calls, demo requests, support tickets, site search and the comparison pages in your category. Add the questions your own pages already answer.
- Group them into topics. Five or six topics that match what you sell, with up to three questions each, is enough to see a pattern. Mark every question as naming your brand or not.
- Ask each question on every engine, on a schedule. Ask the same questions on ChatGPT, Perplexity, Claude and Google AI Overviews, the same way each time, and keep every answer.
- Record four things per answer. Whether you were named, in what position, which brands were named instead, and which pages were cited.
- Call a question won only when the interval says so. Treat a question as one that consistently recommends you when the lower end of the 95% interval on its mention rate is above 50%. Below that, it is contested or lost.
- Sort into won, contested and lost. Won questions need defending. Contested ones need more answers before you act. Lost ones, where a rival is named and you are not, are where the work is.
- Trace each lost question to its sources. List the pages the engines cited and the searches they ran for that question. Mark which mention you. The most cited pages that leave you out are your targets.
- Re-measure after every change. Ask the same questions on the same engines again and compare intervals, not single answers.
What does that look like on a real brand?
The public Tesla example, one scan of 25 buyer questions on four engines, sorted with the rule above.
| Topic | Answers naming Tesla | 95% interval | Call |
|---|---|---|---|
| Supercharger network | 13 of 16 | 57% to 93% | Won |
| Full Self-Driving (Supervised) | 12 of 20 | 39% to 78% | Contested |
| Megapack energy storage | 11 of 20 | 34% to 74% | Contested |
| Powerwall home battery | 8 of 20 | 22% to 61% | Contested |
| Solar panel installation | 6 of 20 | 15% to 52% | Contested |
| Semi truck charging | 3 of 4 | 30% to 95% | Contested |
| SunPower vs Enphase Energy for home battery storage: which is better? | 0 of 4 | 0% to 49% | Lost |
One topic is a clear win: the Supercharger network. Every other topic is contested after a single scan, even where Tesla looks strong, because 20 answers leave a wide interval. The one clear loss is a single question, the SunPower and Enphase head-to-head, where all four engines answered about the two rivals and none named Tesla.
That is the honest reading of one scan: a short list of leads, not a verdict. Daily answers on the same questions narrow every interval. A month of one answer a day gives each question 30 answers on every engine, and a question that names you in two thirds of them has an interval of about 49% to 81% on that engine alone.
Which metrics best measure AI recommendation share and accuracy?
- Mention rate, per question and per engine: answers naming you, out of answers given, with its interval.
- Share of voice: your mentions as a share of all brand mentions in those answers, so you know how many rivals you share the shortlist with.
- Position: where in the answer you are named. First and fifth are not the same recommendation.
- Citations: how often the engines link to your own pages, and which pages they link to instead.
- Accuracy, from questions that name you: whether the engine describes the right company, product and facts.
The formulas are on the methodology page and the terms in the glossary. To put your numbers beside the category leader's, see benchmarking AI visibility against competitors.
How often should I run prompts across different AI assistants?
Often enough that every question collects answers on every engine. Answers move from day to day, so a weekly check sees a different shortlist each time and cannot tell drift from noise. Proofsource's Custom plan asks every tracked question on every engine daily. The free trial runs 2 scans, today and tomorrow, which shows where you stand and which questions to watch, but not yet a trend.
How can I identify the content that drives AI recommendations?
Start from the questions you lose and work backwards. For each one, list the pages the engines cited and the searches they ran before answering, and mark which pages mention you. The pages cited most often that leave you out are your targets: get onto them, correct them, or publish something more useful on the same question.
Proofsource lists the cited pages per question and engine, records the searches each engine reports running, and flags the sub-questions where a competitor or review site is cited and you are not, with the page of yours that should state the answer when it finds one. The citations feature shows how.
From a question list to a list of fixes
A starting set from your site
Proposed from your site, category and rivals, grouped by topic. You edit the list before anything runs.
Named, or who instead
Every question on every engine, with your position, the brands named instead and the pages cited.
One fix per lost question
Each gap tied to the pages that won it, with a drafted fix that waits for your approval.
What buyers ask about finding their questions
How can I check which buyer questions consistently get my company recommended by AI assistants versus competitors?
Ask a fixed set of real buyer questions on every engine, repeatedly, and record per question whether you were named and who was named instead. Call a question won only when the lower end of its 95% interval is above 50%. Proofsource does this on ChatGPT, Perplexity, Claude and Google AI Overviews and lists each question's mention rate per engine beside the brands named instead.
How should I build a representative set of buyer questions to test?
Collect real questions from sales calls, support and site search, group them into five or six topics that match what you sell, keep up to three per topic, and keep questions that name your brand separate. Proofsource proposes a starting set from your site that you edit before anything runs.
Which metrics best measure AI recommendation share and accuracy?
Mention rate per question and engine, with an interval; share of voice against the brands named alongside you; average position in the answer; how often your own pages are cited; and, for questions that name you, whether the answer describes you correctly.
How often should I run prompts across different AI assistants?
Often enough to see a rate rather than one answer. Answers change from day to day, so Proofsource's Custom plan samples daily. The free trial runs 2 scans, today and tomorrow, which is enough to see where you stand but not to call a trend.
How can I identify the content that drives AI recommendations?
Look at the pages engines cite for the questions you lose and the searches they ran first. Proofsource lists both, and flags the sub-questions where a competitor or review site is cited and you are not, with the page of yours that should state the answer where it can find one.
What tools can track AI recommendations for my company versus competitors?
Several AI visibility tools do. Proofsource is one: it asks your buyer questions on four engines, records who is named instead of you, and drafts the fix for each lost question. Our comparison of AI visibility tools sets the main options side by side, with a source for every fact.
Related
Find the questions that leave you out.
See which buyer questions name your brand, who gets named instead, and what to fix. 200 answers. 25 Prompts/Question. Top AI engines. 2 days. Free. No card.
Google AI OverviewsGoogle AI Mode
