What is measured
- A fixed set of customer questions for your business, such as “Who does same-day crowns in Clapham?”, plus branded and comparison questions.
- Five assistants: ChatGPT, Claude, Gemini and Perplexity, asked through their APIs with web search on, and Google AI answers read from a live Google search.
- Your market: each question is asked with your business's country and language.
- Every answer is kept with its provider, model, date, whether your business was named, and the sources it cited.
What a set of results looks like
The table below is illustrative. It uses a fictional clinic and shows the shape of the records, not real answers.
| Provider | When | Named? | Sources cited |
|---|---|---|---|
| ChatGPT | Sample date 1 | Yes | brightsidedental.co.uk/same-day-crowns |
| Claude | Sample date 1 | No | Two other clinics’ pages |
| Gemini | Sample date 2 | Yes | A local directory listing |
| Perplexity | Sample date 2 | Unclear | Named a different branch; marked ambiguous |
Four answers are not a trend. With this many samples, MakeTrend shows the records and says “Not enough evidence” instead of a conclusion.
When a change counts as real
After you publish a fix, MakeTrend compares answers before and after it. It only reports “Improved”, “Declined” or “No clear change” when all of these are true:
- The same provider and model answered in both windows.
- At least twenty matched questions were asked in both.
- Each window has results from at least two dates.
- At least 80% of planned checks completed.
- No more than 10% of answers were ambiguous.
Otherwise it says “Not enough evidence” and tells you why.
What it is not
- Not a recording of the ChatGPT or Gemini apps. Those add memory and personal context that an API test does not have.
- Not a ranking or a guarantee. Answers vary from run to run.
- Not proof of sales. Enquiry records are kept separately, so you can compare them without assuming a mention caused a booking.
Questions
Is this the same as what I see in the ChatGPT app?
Not exactly. Samples come from each provider's API with web search turned on. The consumer apps can add memory, location and other context, so your own screen may differ. Each sample says which provider and model answered.
How many answer checks do I get?
The Business plan includes 1,200 answer checks a month. One check is one question asked to one provider once.
Why does it say “Not enough evidence”?
Because a few answers cannot show a trend. A before-and-after conclusion needs at least twenty matched questions, results on at least two dates in each window, and most checks completing. Until then, MakeTrend says so instead of guessing.
Can I monitor in my own country and language?
Yes. Each question is asked with the business's country and language, and the result records both. The app itself is in English.