A check is one complete measurement pass: every active question on your brand is put to every engine (ChatGPT, Gemini, Google Search, Perplexity and Claude), and every answer is judged and stored. Checks are the heartbeat of brandflare; everything on your pages is built from their results. In the wider industry a check is called a sweep.
How each engine is asked
For the brands you track, ChatGPT and Gemini are asked in their own apps, with web search on and from your brand's market, so each answer is the one a person there is shown, with the sources it cites. Google Search is asked in Google's AI Mode, the answer Google itself gives in search, from the same market. Perplexity answers from live web results and cites them too. Claude is asked as a model, without web search, so its answers come from what the model has learned and carry no sources.
Profiles in the public index are asked through ChatGPT's and Gemini's models, without web search.
The schedule
Every brand is checked once a week, on every plan. Each brand is given its own weekday when it's added, and keeps it. Spreading brands across the week keeps results arriving steadily, and because your brand's day never changes, week-on-week comparisons stay clean.
Between those checks, your visibility is measured every day. On the other six days of the week, the questions that don't name your brand are put to the four engines that search the web (ChatGPT, Gemini, Google Search and Perplexity), and each answer is read for who it names and in what order. Three more engines are measured the same way every day of the week: Copilot, DeepSeek, and Google's AI Overviews, the answer box at the top of an ordinary Google search. That keeps your Visibility, Average position and Flare score current from one day to the next, so a change shows within a day instead of at the next weekly check.
The daily measurement is deliberately narrow: each answer is read for who it names and in what order, and nothing else. Beside your headline figures it fills the cells on Questions & Prompts, where each question that doesn't name your brand shows its latest answer and the last 7 days, and it is counted in Competitors. Tone, accuracy, sources and Brand Narrative come from the weekly check, which reads every answer in full, and alerts are sent after it. Claude answers from what the model has learned, which doesn't change from day to day, so it is asked weekly. Questions that name your brand aren't part of visibility, so they are asked weekly too. Copilot, DeepSeek and AI Overviews are measured for visibility alone: each has a line on your trends, a row under By engine and a column on Questions & Prompts, and no tone, accuracy or sources. DeepSeek answers from what its model has learned, without searching the web; the same question can get a different answer from one run to the next, and asking it every day steadies its figure. Google doesn't write an AI Overview for every question, so your visibility there is the share of the overviews it did write that named you. A brand's extra markets are checked weekly, and when you view all of a brand's markets together the figures are the weekly ones.
Profiles in the public index are checked once a month, each on its own day of the month, on four engines: ChatGPT, Gemini, Perplexity and Claude. Their Flare score is built from the last month of answers, and their changes are measured against a month earlier. A brand you track is checked weekly on its own questions, whatever its public profile does.
The top of each brand page says when the brand's figures were last updated, by its daily measurement or its weekly check, and when the next weekly check is due.
The first check starts on its own the moment you finish adding a brand. It usually takes 20 to 30 minutes, and the brand's page counts the answers in as they arrive rather than showing empty numbers. Recommendations and Brand Narrative are filled in as soon as that first check finishes.
What happens in a check
For each question and each engine, three things happen.
1. Ask. The question is sent to the engine as a customer would ask it, and the full answer is kept word for word. If your brand has a market, the question is asked as someone in that country, and the answer records where it was asked from. When an engine backs its answer with web sources, the cited pages are kept too, which is what fills Sources.
2. Judge. The answer is prose, so reading it is itself an AI task. A frontier-model LLM-as-judge reads each answer for:
- Mentions. Whether your brand is named (under any registered alias or product name), which competitors are named, where each one lands in the answer, and how warmly and how firmly each is recommended.
- Accuracy. Everything the engine says about your brand is compared with your fact sheet, and false claims are raised by severity. Each one is double-checked before it's kept. Things an answer left out are never raised. Details: Answer Accuracy.
- Narrative. On plans with Brand Narrative, the themes in how answers describe you.
3. Store. The answer, its sources, the mentions and any false claims are all saved. Nothing is thrown away: any number on your pages traces back to the exact answers behind it.
Individual answers can fail, because engines have outages and rate limits. A failed answer is left out of your numbers rather than counted as a miss, and a check that's interrupted picks up where it stopped. The same question is never put to the same engine twice in one day.
From answers to numbers
When the answers are in, the results roll up by brand and by engine:
- Visibility: the share of answers that named your brand.
- Average position: where you appeared among the brands named, when you appeared.
- Share of voice: your share of mentions among you and the competitors you track.
- Sentiment: how favourably you were described.
- Accuracy issues: false claims still waiting for review.
These feed the overview, the trends, the Flare score and the engine-by-engine grid on Questions & Prompts.
After the check
Once the numbers are in, alerts go out to the people who've asked for them: a new false claim, a rival passing you, an engine dropping you. You hear about it the same day, not at your next login. A finished check also means a new Export Report can be built.
Why a weekly check beats a spot check
Asking ChatGPT about your brand once tells you almost nothing: the same question gets different answers from one run to the next. Checks solve this with volume and repetition. A full set of questions, five engines, the same day every week, and your visibility measured again every day in between. Single-answer noise averages out, and what remains is real movement in how AI presents your brand, tied to a date, and therefore to whatever you shipped that week. That's the discipline of prompt tracking, automated.
What a check costs you
Nothing per check: they're included in your plan. The number of questions on your plan is what decides how much measurement you get. Fifteen questions tell you whether AI names you at all; forty cover the questions that decide a sale, and make every figure steadier, since each question is another observation. The limits are in plans and limits.