Prompts

This is the page where AI visibility actually starts. Before any score, any citation, any competitor comparison - there's a list of questions you care about, and Trakkr asks AI those questions on your behalf.

If you're new here, read this front to back. If you've been around a while, jump to the section you need from the table of contents on the right.

What is a prompt?

A prompt is a question someone might type into an AI chatbot - exactly the way they'd type it. Things like:

That's it. No keywords, no operators, no formal syntax. Just the question a real human would ask when they're trying to find a product or pick from a category.

You add a prompt to Trakkr. We ask all the major AI models that question. We record which brands they recommend, and where you land in that list. We do it again tomorrow, and the day after, forever. Over time, you get a picture of how AI talks about your category - and whether you're getting recommended.

Prompts are the new keywords

In Google's world, the unit of measurement was the keyword. You ranked for a string of words; people typed that string; pages competed for the top spot.

AI doesn't work that way. There's no ten-blue-links list. There's a generated answer - one paragraph, maybe with a few citations underneath. Either you're in it, or you're not. And the things people type into AI aren't keywords, they're full questions.

So in Trakkr, the prompt is the unit. Each prompt has its own score, its own trend, its own competitors, its own citations. The Prompts page is where you decide which questions matter enough to track.

How Trakkr uses your prompts

Once a prompt is active, here's what happens behind the scenes:

Every active prompt gets sent to eight AI models, once per day, at 3:00 AM UTC.

ChatGPT

Claude

Gemini

Perplexity

AI Overviews

Grok

Meta AI

DeepSeek

For each response, we extract:

• Whether your brand was mentioned at all
• Where you ranked (#1, #2, etc.) if multiple brands were named
• Which sources the AI cited
• Which competitors showed up alongside you
• Sentiment, framing, and the language used about you

That raw data feeds into your visibility score, the trend lines on your dashboard, the citation graph, the competitor comparison, and the audience breakdowns. The prompts you choose are upstream of everything.

A tour of the Prompts page

The page starts with the prompt table, with two more views alongside it: Tags and Audiences.

The rest of this page focuses on the Prompts view. Tags and Audiences have their own deep-dive pages.

The header bar

Above the table you'll find:

• Add - opens the prompt creation modal
• Rerun (Scale and Enterprise only) - manually triggers a fresh report run, limited to 2 per day per brand
• Export - download a CSV of everything visible with current filters
• Manage Tags - opens the tag manager
• Location - if you have multiple markets, the location filter shows here
• Live status - "Last report 4h ago" tells you when fresh data landed, and "23/50 active prompts" shows where you are against your plan limit

The filter row

Just below the header sits a compact filter row with:

• Search - instant filter across prompt text and tag names
• Tags - multi-select pill filter; URL updates so you can share a filtered view
• Audiences - multi-select pill filter (once audiences have been generated)
• Status - All / Active / Inactive
• Volume - 10K+ / 1K-10K / <1K monthly searches

Filters stack. You can search inside a tag, inside an audience, restricted to active prompts. Anything you filter to, you can export.

The table

The Prompts table is a dense, sortable, spreadsheet-style grid. Here's every column it can show:

Click any sortable header to sort. Click again to reverse. The active sort shows a small arrow.

Click the Columns dropdown (or right-click the header) to hide or show columns. Drag column edges to resize. Your preferences persist per brand.

The score and the sparkline, together

The most useful thing in the table isn't any single column - it's the relationship between Score and 7d. The score tells you where you are. The sparkline tells you which way you're moving. Together they tell you whether to celebrate, investigate, or relax.

Adding prompts

You'll add prompts three ways, depending on how much certainty you have about what to track.

1. Add a single prompt you already have in mind

Hit the Add button in the top right, or press N from anywhere on the page. A modal opens. Type your question. Hit Enter.

That's it. The prompt is created and queued for the next research run.

2. Use the Ideas tool to brainstorm

Scroll to the bottom of the Prompts table. You'll see a dashed row that says Explore prompt ideas. Click it to open the Ideas panel.

Ideas is an in-page brainstorming tool that uses AI to suggest prompts you should be tracking but aren't. It has three modes:

Brand mode (default)

Discovery-style category prompts based on your brand's industry, description, and website - the "best X for Y" questions a buyer asks before they know which product to pick. This is the right starting point when your tracking is sparse.

Tag mode

Click any of your existing tag names in the mode pills. Trakkr looks at the prompts already in that tag and suggests more discovery prompts like them. If you have five "pricing" prompts, tag mode might suggest ten more.

This is the mode to reach for when you've identified a theme that matters and want to fill it out properly.

Keyword mode

Type a topic into the input on the right and hit Go. Trakkr generates discovery prompts focused on that specific keyword or angle.

Try things like:

• A feature you want to be known for (enterprise SSO, offline mode)
• A use case you're targeting (remote teams, solo founders)
• A category you compete in (project management, subscription billing)
• A pain point you solve (burnout, scope creep, data silos)

Working with suggestions

Each suggestion appears as a "ghost row" in the table, formatted like a real prompt but greyed out. Hover any suggestion to see why Trakkr recommends it.

• + on a single row adds it instantly
• Checkbox selects multiple; the bulk action bar lets you add them all at once
• Skip removes a suggestion (it won't reappear)
• Regenerate (the circular arrow) refreshes the entire list

Suggestions stop counting against your prompt limit until you accept them. There's no cost to exploring.

3. Bulk discover with Research

Need more than a handful? Use Research instead. Research analyzes ~500 questions in your market in one go, ranked by how often they're actually asked, with full position and competitor data attached. It's the right tool when you're scoping out a category from scratch or quarterly-auditing your coverage.

Writing prompts that matter

You have a finite number of prompt slots. The question isn't "what prompts can I track?" - it's "what are the most important questions I should know the answers to?"

Here's how to choose well.

Start with real customer questions

The best prompts come from things customers actually ask. Mine them:

Cover the customer journey

Discovery prompts come in different shapes depending on how specific the buyer is being. A healthy mix spans the spectrum from broad category questions to narrow "best for my exact situation" ones.

Broad category - "What's out there?"

The prospect is scoping the category. They want to see the headline options.

Use-case specific - "What's best for my situation?"

They've narrowed by job-to-be-done. They want recommendations tuned to their constraints.

Feature or attribute - "What's best at X?"

They have a non-negotiable. They want options ranked by a specific capability.

A good split is roughly 30% broad, 50% use-case, 20% feature.

Keep prompts unbranded

Every prompt should be a category question, not a question about you. The point is to see whether AI recommends you when someone is choosing what to use - if you name your brand in the prompt, you've already given AI the answer.

The first two give you a real signal: did AI list you, and where? The last two are dead-ends - AI is going to talk about the brand you named either way, and "yes/no" answers don't produce the ranked recommendations Trakkr scores against.

Find the specificity sweet spot

There's a balance between too broad and too narrow.

Too broad and you're competing with everyone - you'll never win, and the score won't move. Too narrow and there's no signal because nobody actually asks that question. The middle is where you get useful, movable data.

Common mistakes

Branded or reputation prompts. "Is Stripe good for SaaS?" or "What do users think of Stripe's API?" name the brand for AI, so there's nothing to win and no list of products to score against. Stick to category questions.

Branded comparison prompts. "Notion vs Confluence" might feel high-intent, but it forces AI to talk about both brands - neither one had to earn the mention. The unbranded version ("Best documentation tools for engineering teams") is what tells you whether you actually get recommended.

How-to and educational prompts. "How do I track AI visibility?" produces an explainer, not a product list. There's no ranking, no recommendation, nothing for Trakkr to measure.

Questions nobody asks. "What is the paradigm of agile collaboration in modern subscription billing?" Real people don't ask that. Tracking it doesn't help you.

Near-duplicates. Don't track "best running shoes," "what are the best running shoes," and "which running shoes are best." Pick one phrasing and move on. Phrasing analysis (more below) handles the variant problem more cleanly than spamming the table.

Quick wins

If you do nothing else today, add these three discovery prompts:

1. 1.The headline category question. Best [category] for [your target customer]
2. 2.A use-case prompt. Best [category] for [specific use case or team size]
3. 3.A feature-led prompt. Best [category] with [feature you want to be known for]

That's a baseline that covers broad demand, specific use cases, and the features that differentiate you.

Phrasing analysis

Here's a thing that surprises people: AI rankings can swing wildly based on how a question is phrased, even when the intent is identical.

"Best CRM for startups," "What CRM should a startup use," and "Top CRMs for early-stage companies" might return different brand rankings, even though a human would treat them as the same question. That's a blind spot in your tracking that a single prompt can't see.

Phrasing analysis fixes that. For any prompt, Trakkr generates five semantic variants and ranks each one. You see at a glance which phrasings you win on, which you lose on, and which you're invisible for.

To run it, click Analyze phrasings from the row action menu or hover the score and click the prompt in the popover. Each variant gets a coverage dot:

• Green - you ranked in the top 5
• Grey - you appeared but outside the top 5
• Red - you didn't appear at all (a blind spot)

When you find a blind spot worth fixing, the popover gives you two next steps:

• Track - add that variant as its own prompt
• Diagnose - open it in the Diagnose tool to see why AI didn't mention you

Editing, archiving, and deleting

Inline edit

Double-click any prompt text to edit it in place. Press Enter to save, Esc to cancel.

The detail drawer

Click a row anywhere except the text or checkbox to open the Prompt Detail Drawer - a wider view with the full prompt history, all eight model responses, the citation list, competitor comparisons, and per-model scores. This is where you go when one prompt deserves a deeper look.

Edit modal

Click the actions menu (...) and pick Edit details for the full editor with tag management, audience assignment, and active/inactive toggling all in one place.

Active vs inactive

Toggle the On column to pause a prompt without losing its data.

• Active prompts run daily and count toward your plan limit
• Inactive prompts don't run, don't count, and keep their entire history

Use this for:

• Seasonal prompts (active during holiday season, paused the rest of the year)
• Experiments you want to revisit (turn off after evaluating)
• Anything you don't want to delete but don't want to spend quota on

Archive vs delete

• Archive removes a prompt from the active list but preserves all historical data. You can restore it.
• Delete permanently removes the prompt and everything Trakkr collected for it. There's no undo after the toast disappears.

When you archive or delete, an undo toast appears for about five seconds. After that, the action is committed.

Bulk actions

Pick the prompts you want to act on, then act on all of them at once.

Selecting

The bulk action bar

When prompts are selected, a bar appears at the bottom with these actions:

• Activate / Deactivate - toggle the On state for all selected
• Apply tags - open a tag picker and apply to all
• Move to audience - reassign all to an audience
• Duplicate - copy each one as a new active prompt
• Archive - pause and remove from active list
• Delete - permanently remove

All bulk actions get an undo toast.

Keyboard shortcuts

The Prompts page is built for keyboard users. The shortcuts work whenever you're not actively typing in an input.

Plan limits

Your plan caps how many prompts can be active at once. Inactive prompts are unlimited.

Hit your limit? Either:

• Deactivate prompts you care about less to free up slots
• Add a prompt pack in Settings → Billing to expand your quota

Going further

These related views unlock deeper analysis once you have a healthy prompt list.

Common questions

How many prompts should I start with?

Five to ten. Enough to get a real picture without making the analysis overwhelming. You can add more once you've seen the baseline.

Should any of my prompts mention my brand?

No. Trakkr is built for unbranded discovery prompts - open category questions where AI has to pick which products to recommend ("Best payment processor for SaaS"). Branded prompts ("Is Stripe good for SaaS?") and branded comparisons ("Stripe vs Adyen") name the brand for AI, so there's no listing to win and nothing competitive to measure. Keep every prompt category-level.

How often do prompts run?

Every active prompt runs once per day, at 3:00 AM UTC, across all eight AI models. Scale and Enterprise customers can also trigger a manual rerun with the Rerun button (up to twice per day).

Why is my score different across models?

Each AI model has different training data, different reasoning, and different biases. ChatGPT might love you while Perplexity barely mentions you. That's normal - and it's exactly why we test all eight. The Dashboard shows your per-model breakdown.

Can I import prompts from a spreadsheet?

Yes, but indirectly. Use Research for bulk discovery (it generates ~500 prompts ranked by demand), or paste prompts one at a time via the Add modal. CSV import isn't currently in the Prompts page; ping support if you need it.

What happens to a prompt's data when I delete it?

It's gone, permanently. If you might want it back, archive instead - that preserves the full history.

What if I want to track the same prompt for multiple brands?

Each brand has its own prompt list. There's no shared library - and that's intentional, because brand-specific phrasing usually matters. If you want to track parallel prompts across brands, the easiest way is the Duplicate brand flow in Brands.