Your questions define what brandflare measures. They're the AI era's keywords: the finite, repeatable things buyers ask, against which your visibility is scored. This page covers how your questions are written, what makes them trustworthy, and how to read Questions & Prompts, the section that shows your results question by question, engine by engine. In the wider industry a tracked question is called a prompt, and the grid a prompt matrix.
What a question is here
A question is something a real buyer in your category asks an AI assistant: "best X for Y", "X vs Y", "is X worth it", asking for a recommendation or a comparison. Good questions share two properties:
- They don't name your brand. "Tell me about Acme" tests whether the engine knows you exist; "what's the best project tracker for small teams" tests whether the engine recommends you, which is the one that moves revenue. Questions that do name your brand can be tracked, but they don't count toward your Visibility or your Flare score.
- They're things people actually ask. The set should mirror real buying conversations, not marketing language.
How your questions are written
When you add a brand, brandflare writes its questions from the brand's category, its competitors and its market. A UK brand gets questions a UK customer would ask, in pounds, with the shops and rivals they know.
They are written in English unless you choose another question language in the brief when you add the brand. The choice is only for the questions brandflare writes on that first run: whichever you pick, you can add your own in any language afterwards. Your plan sets how many you can track: 15 on Starter, 40 on Plus, Pro and Agency (plans and limits). The tab row shows how many your plan allows.
The set is yours to shape:
- Add questions. Several at once, one per line, with a live count of how many fit your plan. Repeats are skipped.
- Pause a question that doesn't fit. A paused question stops being checked, and its history stays.
Shape it early, then leave it mostly alone. Prompt tracking only produces meaningful trends on a stable yardstick. If the questions change every week, a rising score could mean rising visibility or merely easier questions, and you'd never know which.
Question type
Every question has a type: the kind of question it is, not its subject — a best-of list, a comparison, a recommendation or one about the brand itself (a leader's types are different, built for the Leadership GEO audit). You can filter the list by type. Questions you add are given one a moment after you save, with Other for the rare one that fits none of the usual kinds. There is nothing to pick.
Type turns forty rows into a handful of stories: you're strong on best-of questions and absent from comparisons is a finding you can hand to someone.
Branded and unbranded
A branded question names your brand: "Is Acme any good?". An unbranded question doesn't: "What's the best project tracker for small teams?". brandflare works out which is which from the question's own words, using your brand's name and the other names registered for it. A question in another language is read for it when you add it, because your brand's name may be spelt another way there.
The two tell you different things:
- Unbranded questions show whether AI thinks of you. Visibility, Share of voice, Average position and All engines are measured on these alone.
- Branded questions show what AI says about you to someone who asks directly. You're nearly always named, so what matters is how you're described and whether it's true.
Branded vs unbranded on your overview sets the two side by side on the latest answers: how often you're named, named first and recommended, the tone, and the share of answers that carry a false claim. In Questions & Prompts, the tabs show All your questions, the Branded ones or the Unbranded ones, and each branded question carries a Branded tag.
Questions in other languages
A question is asked in the language it is written in. The questions brandflare writes for a new brand are in English, or in another language if you choose one as the question language in the brief when you add the brand. Whichever they are in, you can add your own in any language your customers use: type the question as they would, in their words and their script. A brand can hold questions in more than one language, and a brand's extra markets ask the same questions in the same language.
- The engines answer in that language. Each answer is read as it was given, not translated first.
- Everything brandflare writes stays in English. A question shows what it means in English underneath. The brands an answer named, how the answer describes you and any claim to review are in English, with the answer's own words kept beside a claim.
- Your brand is counted however it is spelt. An answer in Arabic may write your name in Arabic letters, and in more than one way. It counts as your brand each time, and each spelling is highlighted in the answer.
- It counts like any other question. A question that doesn't name your brand counts toward Visibility, Share of voice and Average position, and is asked every day.
The grid
Questions & Prompts lays your results out as a grid: one row per question, one column per engine. ChatGPT, Gemini, Google Search, Perplexity and Claude are asked every question. Copilot, DeepSeek and AI Overviews, which are measured for visibility alone, are asked the questions that don't name your brand, so their cells are empty on a branded row. Each cell answers the smallest question there is: did this engine's latest answer name your brand, and where?
- A cell shows 1st, 2nd and so on: your place among the brands the answer named. Being the first brand recommended and being the seventh name in a comparison table are both mentions, but they're very different outcomes.
- Or it shows Not named.
- On AI Overviews a cell can show No overview: Google wrote none for that question that day, so there was nothing to be named in.
- Named by counts how many engines name you on that question.
- Each question also says how often it's asked and what kind of question it is.
Questions that don't name your brand are asked every day, so their cells show the day's answer. Under each of those cells is a row of dots, one for each of the last 7 days: a filled green dot for a day you were named, an amber ring for a day you weren't, and a faint mark for a day with no answer. Seven green dots is a question you hold. A mix is a question on the edge, and the dots tell the two apart where a single answer can't. Claude, and questions that name your brand, are asked weekly and have no dots.
A key above the table explains the cells. The tabs cut the list to All, Branded or Unbranded questions, and the rows under them to a type. The weakest questions come first, then the most asked, and paused questions sit at the end.
Reading an answer
Every cell is backed by a stored answer. Open one and you read the response as the customer would have seen it, with headings, lists and tables intact, your brand highlighted and competitors underlined. An answer in another language has Translate to English: press it once and the answer can be read in either language from then on. The translation is for reading; the answer is measured as the engine gave it. Beside it are the brands the answer named, in order, with how it described each, the websites it used, and where it was asked from. A daily answer is read for who it names, so it lists the brands in order without how each was described; that comes with the weekly answers. Nothing in the grid is more than one click from its receipts.
Reading the grid well
This is where a headline number stops being a number and becomes a diagnosis. Patterns worth looking for:
- Empty rows: questions where no engine names you. These are your losses, and your content roadmap: each is a buyer question your material doesn't currently win. They sit at the top of the Unbranded tab.
- Empty columns: one engine that leaves you out while others don't. Engines disagree because what they've learned and what they retrieve differ; a lagging engine tells you where your footprint is thin. It also drags Consistency in your Flare score.
- Named, but last: rows where you appear in a poor position every time. You're in the consideration set but framed as an afterthought. These often respond to work faster than empty rows do.
- Flickering cells: dots and rings mixed under a cell, named one day and absent the next. You're on the edge of the engine's answer, and small improvements in the sources it draws on can tip these your way.
Because engines give different answers from one run to the next, treat a single change as a hint and a direction that holds across checks as a finding.
How the grid feeds everything else
Each cell is one observation. Visibility is the share of cells where you were named; Average position comes from the places; All engines counts the unbranded rows where every engine named you. Competitor mentions come out of the same answers, which powers share of voice and the head-to-head in Competitors. The missed questions feed What to do next on your overview, Recommendations, and the plays in Export Report. Cells for questions that don't name your brand refresh every day, and the rest with every weekly check.
Fifteen questions or forty
Fifteen questions cover your core buying questions. Forty add breadth, with more sub-niches, use cases and comparisons, and correspondingly steadier figures, since every extra question is another observation averaging out the engines' noise. If your category is broad, the number of questions is usually the most consequential difference between plans.