Platform · Prompt Tracking
Every day, we ask AI the questions your buyers ask, and read every answer.
A prompt is one of those questions, in the buyer’s words, with no brand in it. We run each one across ChatGPT, Gemini, Perplexity, Copilot, and Google AI Mode once a day, then keep what comes back: the answer, the sources it cited, and which brands it named and where. Everything else in the platform reads from those answers.
What a prompt is
A prompt is a question a buyer would type: “best CRM for a small real-estate agency”, not “is Acme any good”. It never names you or a competitor, so any brand can appear in the answer, or fail to. Each has one country, because New York and London get different answers.
Prompt tracking is the foundation of the platform: competitor share, persona splits, all 24 report metrics, and the routines that write for your weakest questions all read from these answers. Get the prompt set right and everything downstream measures something real.
How it works
Once a day, every prompt goes to every engine
Each workspace has a fixed hour. When it comes round, every active prompt goes to each engine you track, one job per prompt per engine, through Queued, Answering, Processing, and Completed. The picture refreshes once a day rather than in real time, and a day’s set can take minutes or hours to settle.
For every completed answer we keep the answer text, with the model that wrote it and when; every source it cited, with its position; and, for your brand and each competitor, how many times it was named and how early the first mention appears.
Detection is plain text matching, not a model’s opinion: each brand’s name variants and owned domains, looked for in the answer and the citation hosts. A name is either in the text or it is not, and nothing is scored for sentiment. Change an alias and every past answer is re-read.
Three numbers
Mention rate, share of voice, and position, read together
Each answers a different question, and where two of them disagree is usually what needs work.
Mention rate
The share of answers that name you at all. Thirty answers, sixteen name you: 53%. That is your headline visibility number, and the first to move when new content lands.
Share of voice
Your mentions as a share of every mention across you and your tracked competitors, counted by occurrence: an answer that names you three times adds three. The share is zero-sum, so a competitor gaining is you losing.
Position score
How early your first mention appears, 0 to 100, averaged over the answers that name you. 100 means you open the answer, and a low score means you turned up as an afterthought. It is not a rank against your competitors.
Why it matters
The number that tells you whether the work is working
Because every prompt runs every day, a change on your site shows in the answers within days, not quarters. Ship a comparison page on Tuesday and watch that prompt’s mention rate move, engine by engine, against the competitor named instead of you.
Mention rate and position score together put each prompt in one of four quadrants, split at 50% and 50 points. Healthy: named often, named early. Improve position: named often, but late. Improve visibility: named early when you appear, but rarely. Weak: neither.
That is how we decide what to write next: the weakest prompts, and the ones a competitor wins by the widest margin, become the roadmap.
Where prompts come from
Chosen with you, grown from real demand, replaced rather than reworded
Fifteen come out of onboarding, five per buying stage, reviewed with you before the first run. From there the set grows through research, and a prompt that has already run is replaced rather than edited.
Prompt research from keywords or URLs
Feed it up to 20 keywords you already target, or up to 5 URLs, yours or a competitor’s, and it returns topics with real search demand, each with unbranded candidate prompts. Anything too close to a prompt you already track is flagged Already covered; we review the rest with you and promote the keepers.
Locked once run, replaced instead of edited
After its first run a prompt’s text and country freeze, because every metric reads them live. Rewording it would relabel months of answers as if they had always been about the new question. So we replace instead: the original is archived with its history, and a successor starts its own, with the same tags and persona.
Inside our GEO work
Where prompt tracking sits in the method
It measures all three pillars, and every engagement starts here.
01 · DISCOVERThis feature
The prompt set is how we find out where you show up and where you are missing, engine by engine, before anything else is decided.
02 · OPTIMIZE
Every fix and every article is aimed at a specific prompt, and the next day’s run is how we know whether it landed.
03 · DETECT
The same prompts, every day, so a shift in who the engines recommend shows up as a movement in the series before it reaches your pipeline.
Questions about prompt tracking
What clients ask about the prompts
How are the prompts chosen?
At onboarding we read your site and propose fifteen unbranded, evergreen questions, five for each buying stage, then go through them with you. After that, new prompts come from Prompt Research, which starts from keywords or URLs and real search demand, or from your team when a question comes up in sales or support. A person reviews every prompt before it is tracked.
Do you track my competitors’ names in the same answers?
Yes. Detection runs once per answer for your brand and every competitor in the workspace, using each brand’s name variants and owned domains. That is where share of voice and the competitor rate come from. Add a competitor later and the answers already collected are re-read for it.
Can I add my own prompts?
Yes. Add one in the app, or through the MCP server if your team uses AI agents, set its country, and give it tags and a persona. Active prompts count against your workspace’s allowance; archived prompts do not. Keep it unbranded: a prompt that names you cannot fail to mention you, so it cannot measure anything.
Why can’t I edit a prompt after it has run?
Because every metric reads the prompt’s text and country live. If you reworded it, every answer already collected would be reported as if it had always been about the new wording, and the trend would lie. Instead you replace it: the original is archived with its history, and a successor starts fresh with the same tags, persona, and topic volume.
What does position mean?
How early your first mention appears in an answer, scored from 0 to 100, where 100 is the very first words. Because it is averaged over the answers that name you, an answer with no mention lowers your mention rate rather than your position. It is not a ranking against competitors; the Position Quality report shows who tends to be mentioned first.
Built on these answers
Where these answers go next
- The same answers read for every competitor you name: share of voice, who is mentioned first, who is gaining.
Competitor Research
- Saved views over the prompt data: any engine, country, tag, or persona, ready for the monthly call.
Custom Reports
- A standing instruction to write about your weakest prompt, chosen from Prompt Performance, for a person to approve.
Content Routines
- Get named when a buyer asks AI which tool to buy: the questions SaaS buyers ask, and the comparison pages the engines quote.
B2B SaaS
See what AI is already saying about your brand.
A free audit shows you exactly where you stand across ChatGPT, Gemini, Perplexity, Copilot, Claude, Grok, and AI Overviews, and what it would take to close the gap.