Research
Ask around 500 generated questions about your category once each, then promote the ones worth watching onto your tracked prompts.
Research generates around 500 about your category, asks each one once, and records which brands the answer named and in what order. It is the wide sweep you run to find out what is worth tracking: Prompts is the daily monitor for the short list you chose from it, and Diagnose is the deep read on a single question.
Open it when you set up a brand, after a competitor, launch, or rebrand changes the field, and every quarter or so to catch drift. It is not a page you check daily. A full scan can only be refreshed once a week, and nothing on it moves in between.
Running a full scan
Press Run Research. It takes about five minutes.
Three things about the sampling are worth knowing before you read the numbers.
- The questions are written fresh for the run. Trakkr generates them from your brand description, industry, competitors, and website, spread across intents (recommendation, comparison, best-for) and focus areas (general, budget, enterprise, features, integrations). They are a separate set, not your tracked prompts. A thin brand profile produces generic questions.
- Each question is asked once, of one model. A scan is built for breadth, so it takes a single pass per question. Your tracked prompts are the opposite: every one of them runs against all eight models, every day. That is why a Research position and a Prompts score for the same wording rarely match.
- It is a snapshot. The page shows your latest scan only. Running again replaces what is there. Anything you want to watch over time has to move to Prompts.
The Refresh button appears on the Monday after your last run, so a full scan is roughly a weekly ceiling rather than something you can retry on the spot.
Reading the results
Three blocks sit above the question table. Read them in order and a shape usually emerges before you reach the rows. Focus Mode collapses the stats and leaders panel when you want the table on its own.
The insights strip
A row of chips calling out slices where your result diverges from the rest of the run. Trakkr does the statistics first, then writes each card from those exact numbers, so a chip is never a hunch. The strip needs at least 30 questions in the run before it shows anything.
| Chip | What it means |
|---|---|
| Strength | A slice where you beat your own baseline. |
| Weakness | A slice where you fall below it, usually a coverage gap. |
| Opportunity | Questions where you are close to the line but not over it. |
| Topic | A theme worth knowing about, neither strong nor weak. |
| Competitor | A rival winning one slice. Normally the most actionable card. |
| Pattern | A signal that did not fit the other buckets. |
Open a card and use Show in table to filter the rows behind it.
The stat row
Four numbers set the scene: Analyzed (questions the scan covered), Appearing In (how many named you, with the percentage), Top 3 Positions, and Opportunities (questions where you were not in the top 10).
Market leaders and the position donut
Market Leaders ranks the brands the run mentioned most, each with a bar and your own count beside them. A name you do not recognize sitting at number two is worth chasing down.
Your Position Distribution splits every question into Top 3, #4-10, and Missing. Click a slice to filter the table to it.
| What the donut looks like | What it usually means |
|---|---|
| Mostly Top 3 | You are an established pick. Defend rather than chase. |
| A big #4-10 slice | AI knows you but does not lead with you. The most movable position. |
| A big Missing slice | AI either does not know you or does not recommend you. The largest lever, and the slowest. |
| Split evenly | Different corners of the category treat you differently. Go to Topics. |
The question table
One row per question. Rank is your position in that answer, AI Vol is the estimated monthly volume for the question (a range, with a hover card carrying the estimate, its confidence, the platform split, and AI Overviews exposure), and Prompt is the question itself. All three sort. The search box filters by text, and the pills filter by position: All, Top 3, Close (positions 4 to 10), and Missing (not in the top 10). Click a row to expand the ranked list of brands the answer gave.
| Rank | Read it as |
|---|---|
| #1 | The answer leads with you. |
| #2-3 | In the headline recommendation. |
| #4-7 | Named, but not prominent. |
| #8-10 | Buried at the end of a long list. |
| - | Not named at all. |
Turning a scan into tracked prompts
A scan is only useful once something leaves it. There are two ways out, and an export.
Quick Wins is the fast one. The button in the header carries a count of untracked questions where you placed 4th to 7th, nearest first. Those are the positions where the least work moves you furthest. Filter to 4-5 or 6-7 inside the modal, select what you want, and track them in one action.
Picking by hand is the deliberate one. Every untracked row has a Track button, checkboxes let you take several at once, and rows where you were not named carry a Diagnose button that opens that question in Diagnose to explain why. Right-clicking a row gives the same choices plus Show competitors.
Export writes whatever the current filter shows to CSV, including position, volume, topics, and the competitors named.
Whatever you track joins the daily pipeline from the next run onwards. The scan itself stays a one-off.
Topics
The Topics tab appears next to Full Research once you have a scan, with your remaining deep scans counted beside it. It looks at one slice of the market instead of the whole thing: enterprise pricing, open-source alternatives, a vertical you are moving into. Every row is one theme, at one of two depths.
- Estimated rows are rolled up for free from the topic tags on your last full scan. Instant, no cost, and only as good as the questions that run already covered.
- Deep scans are real measurements. One deep scan generates 50 fresh questions about a single topic and takes about three to five minutes.
Each row shows the Topic, your Mention rate, your Avg pos, and a Visibility bar, on the same scale as a full scan so the two are comparable. A Deep badge marks the rows that were actually measured. Sort by any column to find your best and worst themes.
To add one, press Add topic, name something narrower than your market, add an optional line of context to steer it, then Run deep scan. Trakkr offers suggestion chips drawn from your own market, and a rail called Topics from your research you haven't explored surfaces themes the scan found that you have never zoomed into.
Opening a topic gives you its Run history (deep scans accumulate, so a theme builds a timeline you can click through), Sub-topic strengths, Market leaders + position distribution scoped to that topic, and Best / missed prompts. This is the one durable record in Research: a full scan is replaced each time, while deep scans are kept per topic.
The allowance is per active brand, so it scales with the brands you actually track, and it resets on the 1st of the month. A brand has to be active for a deep scan to run on it.
| Plan | Deep scans per month, per active brand |
|---|---|
| Growth | 5 |
| Scale | 5 |
| Trial | 1 |
Common questions
How does Trakkr decide which questions to generate?
It reads your brand description, your declared industry and competitors, and your website, then asks a model to write 450 to 550 discovery-style questions spread across a mix of intents and focus areas. The quality of the run tracks the quality of your brand profile, so fill that in before your first scan.
Are these the same as the prompts I track?
No. A scan writes its own independent set for that run. It does not read from, change, or consume your tracked list, and it does not use prompt slots. The only crossover is the one you make by pressing Track on a row.
Why is my Research position different from my Prompts score?
They measure different things. Research asks each question once, of a single model. Your tracked prompts run across all eight models every day and average. Expect a question's number to move once it starts running on the full fleet.
What is the difference between an estimated topic and a deep scan?
An estimated topic is a free roll-up of questions your last full scan already asked, grouped by their topic tags. A deep scan asks 50 new questions written specifically for that topic. Estimated tells you roughly where a theme sits, a deep scan measures it.
Can I keep an old full scan?
Not on the page. The full scan view shows the latest run only, and refreshing replaces it. Export to CSV before you refresh if you want the rows, or deep-scan the slice you care about in Topics, because those are kept and build a per-topic history.
Why did Quick Wins suggest a question I would never track?
Quick Wins ranks purely by how close you were, not by whether the question fits what you sell. Any untracked question where you placed 4th to 7th qualifies. Treat it as a shortlist to prune, not a checklist to clear.