We ask what your buyers ask.
Before a buyer ever calls you, they ask an AI who to use. Sentinel asks the same questions, week in, week out, and records who gets recommended. When it is not you, you find out, and you find out what to do about it. Two minutes on this page and you will know exactly how it works.
From your buyer's question to your score.
We ask real buyer questions
"Who should we use for warehouse fit-outs in Bristol?" Not keywords. The questions people with budget actually ask.
The AI engines answer
ChatGPT, Claude, Gemini and Google AI Overviews, asked the way a real buyer would ask, several times over, on a schedule.
We count who gets recommended
Recommended by name, not mentioned in passing. You and your rivals, counted across every answer.
You get a number that moves
A score out of 100, tracked over time, with the questions you are losing and a ranked list of what would win them back.
Three reasons the number holds up in a boardroom.
Asked like a buyer
We see what a logged-out buyer sees. No special access, no accounts that flatter the answers, nothing your customers would not see for themselves.
Repeated, not screenshotted
One answer can be a fluke. We ask again and again, on a schedule, and the trend is what you act on. One screenshot is weather; the score is climate.
The receipts are kept
Every run is recorded. If a number ever looks odd, you can see the actual answers behind it. Ask any other tool to show you that.
What CEOs ask us, answered straight.
I asked ChatGPT myself and it named us. Is your score wrong?
Neither is wrong. The engines answer differently by day, phrasing and account, which is exactly why we ask repeatedly on a schedule instead of trusting any single answer, yours or ours. The trend across hundreds of asks is the truth. One screenshot, in either direction, is a good day or a bad one.
What if the score says we are fine?
Then you have two things most of your rivals do not: proof, and an alarm. Strong positions get defended, not just built. You will know the day a competitor starts taking the questions you currently win.
How current is it?
Free accounts refresh monthly, paid tiers weekly or daily. The engines themselves change constantly as the models update, which is why a one-off audit is out of date the month after you pay for it.
Who decides which questions get asked?
We build each category's set from how real buyers actually ask, and we keep it steady over time so this quarter compares honestly with last. You can add your own questions on any plan, including free.
Do I need to understand AI to use this?
No. You look at one number, its direction, and a short list of what to do next. Your team can go as deep under the surface as they like.
The technical detail, for whoever you send this page to
The question set. Each category's questions are held fixed and version-controlled, so scores compare like with like across quarters. Client-added questions are scored separately from the shared set.
Counting recommendations. We distinguish a named recommendation from a passing mention, match brand names properly (including variants and misspellings), and score share of voice against named competitors, with sentiment recorded.
Sampling. Every question is asked multiple times per engine per cycle, mirroring a logged-out buyer. Variance is handled by repetition and recorded distributions, never by selecting favourable runs.
The evidence trail. Raw answers are stored per run. Any rendered figure can be traced to the responses that produced it, and anything estimated or sampled is labelled on the surface where it appears.
"If you ever showed me an invented number, I would cancel and tell everyone."
Quite right. So Sentinel is built so it cannot happen: no recorded answer, no number. That is the whole quality bar, and it is why the score is safe to put in front of a board.
See where you stand, then judge us.
If the number is good, you will sleep better. If it is not, you will know exactly what to do next. Either way, you stop guessing.
Run your free check →