Prompts

A prompt is a question someone might type into an AI chatbot, written the way they would actually type it. Trakkr records who the tracked AIs recommend for your active questions.

That makes the prompt the unit everything else is built on. Your visibility score, your competitor comparisons, your citations, and your audiences all come from the answers these questions produce. Open Prompts when you want to change what Trakkr measures.

What a prompt is

No keywords, no operators, no syntax. Just the question a person asks when they are choosing what to use.

Open category and buyer questions make the clearest visibility baseline: the model has to choose which brands to recommend. The tracking collection asks for a ranked list of brands for the question, rather than reproducing a full consumer chat. Trakkr can also collect branded questions and apply the same position scoring to them, so choose the prompt set with that behavior in mind.

How Trakkr scans your prompts

Daily tracking needs an active trial or paid subscription. Check the report date and each model's available results before comparing scores. A completed report does not guarantee a usable result from every provider.

• ChatGPT
• Claude
• Gemini
• Perplexity
• AI Overviews
• Grok
• Meta AI
• DeepSeek

Trakkr records the ranked brands in the collected results and sources where the provider returns them. Your position determines visibility; sentiment is a separate measurement.

The Prompts page

Four tabs sit under the page title.

The rest of this page covers the Prompts tab. Tags and Audiences have their own pages.

The header

Add prompts is the main button. Its menu holds Import a file (a CSV or sheet) and Prompt ideas, which opens the Ideas tab. Export downloads the prompts in the current view as CSV (a paid feature). Set health opens the review of your whole set, and Wording sets how prompts read in each market once a brand tracks more than one. A market filter appears under the title once there are several.

Filter and Display

The view pills above the table pick the list: All (the default), Moved (prompts whose change left their usual range), Needs attention (low scores and prompts losing ground), Blind spots (prompts where you are effectively invisible), and Paused. Saved views of your own sit after them.

Filter narrows the rows by tag, audience, topic, and AI volume (10K+, 1-10K, under 1K). Picking two tags matches prompts with either one, not both.

Display decides how the rows read. It holds the columns, the sort (Score, Change, Volume, Date added, and Google measures when Search Console is connected), and Group by (Nothing, Tag, or Audience). The search box also understands words like "falling", "paused", or an AI name and offers a filter chip for them.

The columns

Display lists the 7D column as 7 days and the On column as Tracking on or off. Right-click a header to sort it, hide that column, reset its width, or open the full column list. Drag a column edge to resize.

Health

The Set health button in the header carries a score out of 100 and a band: Strong, Good, Fair, or Needs work. Open it for the full review, which names what your set does well, which prompts to prune, which to add, and the specific problems it found (duplicates, branded prompts, questions too narrow to produce a list). It needs a paid plan and at least three prompts, and it recomputes when your set changes.

Adding prompts

Select Add prompts to open the composer above the table. Type one prompt and press Enter, or paste a list and every line lands at once. Import a file (or dropping a file on the composer) takes a .csv, .tsv, or .txt up to 2 MB and lets you map its columns. Tags go after a pipe or a tab, so Best CRM for startups | pricing | discovery creates the prompt with two tags, and two columns pasted from a spreadsheet work the same way. Nothing is written until you select Track, rows you already track are marked, and you can stage up to 500 at a time. Press Tab on a draft to see other ways to ask it.

The Ideas tab lists questions worth tracking that you do not track yet, grouped by topic, each with the answer it got where Research asked it. Select Track on a row, or select several to track them in one go, or Dismiss the ones you do not want. Need ideas? in the composer pours ideas in as drafts. Nothing counts against your limit until you track it.

When you need more than a handful, use Research instead. It works through hundreds of questions in your market at once, ranked by demand and already carrying position and competitor data.

Writing prompts that matter

Choose useful prompts

You have a fixed number of active slots, so the question is which answers you most need to know.

The best ones come from questions customers already ask. Support tickets, sales call notes, your FAQ page, Search Console queries, community forums, and review sites all beat a brainstorm, because the wording is real.

Cover the range, from scoping the category to naming a hard requirement.

There is a sweet spot between too broad and too narrow. Too broad ("running shoes") and you compete with everyone, so the score never moves. Too narrow ("best running shoes for a flat-footed 38-year-old training a first sub-3 marathon") and nobody asks it, so there is no signal. "Best running shoes for marathon training" is the middle that moves.

Keep research questions separate from the discovery baseline. A branded comparison or how-to may be useful to investigate, but does not make the same test of unaided recommendations. Avoid wording that does not match any real buyer need.

Use this short selection pass:

1. 1List the buyer questions from sales, support or search data that lead to a purchase.
2. 2Choose a few category, use-case and requirement questions. Write each in the language and market you want to measure.
3. 3Put branded research questions in a separate tag, if you need them. Use the tag filter when reading the discovery baseline.
4. 4Select Add prompts, stage the questions and review duplicates before saving. Check that tracking is on for each one.
5. 5Read the first usable results for each model. Keep useful zero-scoring questions: removing them would hide the gap.
6. 6Keep the set stable for comparisons. Record additions, wording edits and deactivations because they change what the score measures.

Near-duplicates are another trap. Tracking "best running shoes", "what are the best running shoes", and "which running shoes are best" spends three slots on one question. Pick one and let Phrasings handle the variants.

Phrasings

Rankings move with wording. "Best CRM for startups" and "What CRM should a startup use" can return different brands, even though a person reads them as the same question. A single prompt cannot see that.

Phrasings asks five other wordings of a prompt and ranks each one. Open the prompt, find Other ways to ask it on its page, and select Ask other wordings. It needs a paid plan. Each wording is asked once, of one AI, so a row carries one answer square and the place you landed, plus who was named first.

A wording that leaves you out gets two actions on hover. Track adds it as its own prompt. Diagnose opens it in Diagnose to find out why the model skipped you. The foot of the section tracks every wording that leaves you out in one go.

The prompt drawer and page

Select a row to open that prompt in a drawer beside the list. It carries the score with its change, one row per AI with your place, who was named first, and the pages the answers read. The foot offers Diagnose this prompt. The head has Full page (or hold Cmd, Ctrl, or Shift and press Enter) for the prompt's own page, with the score and its trend chart, every answer, the sources behind them, the other wordings, and a rail with the prompt's details and recent activity. Ask Agent is in the head too.

J and K (or the arrow keys) walk the same filtered order you left the table in. Esc closes the drawer.

Editing, pausing, and removing

Choose Edit from the row menu, or press E, to open the editor with the text, tags, and market together.

Pausing is the control that matters most. An active prompt is included in scheduled tracking and uses a slot. A paused prompt is left out of future tracking, gives the slot back, and keeps everything it has collected. Use Pause (or P) for seasonal prompts, for experiments you want to revisit, and for anything you would rather not lose. The Paused view lists them, and Resume brings one back.

A prompt's market

A prompt is asked in every market you track unless you give it a market of its own. Choose one under Market in the row menu or the editor to ask that prompt in one place only:

• If you track that market, the prompt is asked and reported there, using that market's wording.
• If you don't, it is asked from the country you chose, exactly as written, and kept with your primary market's results.

Every model gets the prompt's country, not your primary market's. In the table the prompt shows its flag, and a market that doesn't ask it shows no score for it.

Bulk actions and keyboard shortcuts

Click checkboxes to select, or press X on a focused row. Shift-click extends a range. The bar at the bottom carries Tag, Audience, Pause (or Resume), Move to Accuracy, Delete, and Ask Agent about these. Right-click a row for its menu, which also has Duplicate, Copy text, Diagnose, and Create content.

Shortcuts stay quiet while you are typing in a field or a dialog is open.

Plan limits

Growth and Scale normally include 50 active prompts per brand. Inactive prompts do not use active slots. Add-ons and managed accounts can change the allowance; use the limit shown on your Prompts page for your account.

Enterprise caps are set in the contract.

Common questions