For the complete documentation index, see llms.txt. Every page on this site is also served as Markdown: append `.md` to any URL, or send `Accept: text/markdown`.
Skip to content

Measure · Prompt tracking

Track the questions your buyers ask AI, and where you stand on each

Prompt tracking reads the questions your buyers ask AI, worded the way they type them, every day on each engine you track: ChatGPT, Perplexity, Claude and Google AI on Starter. Each prompt shows whether you were named, linked or beaten by a rival, with n beside every rate. The crew proposes new questions weekly; you choose what is tracked.

7 days free · card requiredHow the trial works

Analytics · Prompts · By promptattensira.com · 61 prompts
Standing · named in
what is Attensiran = 60 · linked in 95%
Won · 100%
Attensira vs Profoundn = 60 · linked in 93%
Won · 100%
AI search visibility softwaren = 109 · Semrush named in 58%
Lost · 0%
AI prompt volume tooln = 105 · Profound named in 58%
Lost · 0%
AI crawler logs analysis tooln = 105 · Ahrefs named in 17%
Lost · 0%
Researcher agentTrack these questions?
“12 Best Scrunch AI Alternatives for B2B Teams”: 272 AI crawler hits in 14 days, no prompt tracking it

Choices: Track · Dismiss

Shown as a baseline, not a result.

Researcher agentResearcherattensira.com workspace, read over the Attensira MCP · · 30 days to 2026-10-04, 24 of 31 days sampled
  • Researcher agentResearcher · Proposes up to three new buyer questions a week, each with the evidence behind it.

  • Analyst agentAnalyst · Reads every answer and labels where you stand, per engine and across them.

What does the crew do with this reading?

The documented flow, step by step. It is an example: no brand, no IDs and no numbers.

See how the crew fixes this: Content optimization
“A question your buyers ask”Example
  1. Read the question on every engine your plan tracks, once a day per country.
  2. Labelled each engine's answer: named and linked, named, lost to a rival, or absent.
  3. Kept every answer word for word, with its sources, as the evidence.
  4. Checked the result against the last window before calling anything a change.
  5. Filed what it found on the prompt's row, and in your Inbox when it needs a decision.
  6. Handed to Content optimization: the page that should have been quoted.
Last stateHanded to Content optimization
Example business · not a customer

How does prompt tracking work?

A prompt is a question a person types into an assistant, not a keyword. The answer is prose, so what is measured is who it names and links.

  1. Start with about fifty

    At signup the crew chooses your first prompts and files them into four to eight topics, so a new workspace never opens on an empty grid. You confirm or change every one.

  2. Read daily, per engine

    Each prompt, engine and country is read at most once a day. Starter takes one answer per reading; Growth and up take three, so a rate is steadier. A failed read is left out, never counted as a zero.

  3. Label where you stand

    Each engine gets a cell: named and linked, named, lost to a rival, absent, or not measured. Across engines the prompt is Won, Contested, Lost or Absent, sorted losses first.

  4. Find the questions you miss

    Weekly, the crew looks for buyer questions you do not track yet: URLs rivals are cited on, and pages AI crawlers fetched with no prompt behind them. At most three arrive as one Inbox row.

What does prompt tracking show you?

  • Track questions, not keywords

    A prompt is the question a buyer types into an assistant, worded the way they type it. The answer is prose, so what is counted is who it names and links.

    Prompts in the docs
    attensira.com · ChatGPT and Perplexity · 30 days
    • Questions that name us308 of 311
    • Questions a buyer asks without us2 of 1,481

    The second row is the honest test, and it is our baseline.

    attensira.com's own workspace · read 1 Nov 2026 · n=1,792 answers
  • Label where you stand on every engine

    Each engine gets a cell: named and linked, named, lost to a rival, absent, or not measured. Across engines a prompt is Won, Contested, Lost or Absent, losses first.

    Prompt state in the docs
    Prompt · standing by engine
    • ChatGPT · Lost to a rival
    • Perplexity · Named
    • Claude · Absent
    • Google AI · Named and linked
    Example business · not a customer
  • Group prompts into topics

    A topic pools every successful read of its prompts, so it carries a signal one prompt cannot. The crew files your first prompts into topics at signup, and you confirm every one.

    Topics in the docs
    Topics
    Pricing questions
    Pooled over every prompt in it
    Comparisons
    Pooled over every prompt in it
    How-to questions
    Pooled over every prompt in it
    Example business · not a customer
  • Find the questions you are not tracking

    Weekly, the crew looks for buyer questions you do not track yet. At most three arrive as one Inbox row, and approving it adds them the way you would by hand.

    Prompts in the docs
    Inbox · Suggested prompts
    1. Found URLs rivals are cited on that no prompt of yours covers
    2. Found pages AI crawlers fetched with no prompt behind them
    3. Filed three suggestions as one Inbox row
    Example business · not a customer

Where do new prompts to track come from?

From your own AI crawler traffic and your rivals' citations. Once a week the crew proposes questions, and you decide which take a slot. This is the proposal on our own workspace.

Inbox1 waiting for you
  • Researcher agentApproveTrack these questions?

    Weekly · since 28 Sep

    “12 Best Scrunch AI Alternatives for B2B Teams in 2026”: 272 AI crawler hits in 14 days. “How to verify an AI crawler is genuine”: 11. “12 Best Writesonic GEO Alternatives”: 10. The plan has 105 prompt slots left.

attensira.com Inbox (role shown is the one whose tool makes the item), read over the Attensira MCP ·

Nothing ships without your yes, unless you let a lane ship on its own.

One workspace setting decides how far the crew goes on its own. Draft is where every workspace starts.

  1. Ask

    Stops and asks before anything leaves the workspace. A card shows exactly what it wants to do.

  2. DraftDefault

    Does the work inside the workspace on its own and files everything else for your yes. Nothing is published or sent until you approve it.

  3. Act

    Carries safe changes through on its own: structured data, meta tags, alt text, internal links and heading levels. This is how you let work ship on its own.

  • Claims always wait

    A number, a price, a product or competitor name or a commitment waits for your yes in every mode, Act included. A change you edited before approving always counts as one. In the docs

  • Unanswered approvals expire after 7 days

    If nothing is decided in that window, the run closes without doing the thing and files a row in your Inbox saying so. In the docs

  • Every act has a receipt

    One dated row per act, with its status. Failed attempts are listed, not hidden, and a withdrawn one is marked cancelled, never removed. In the docs

  • You can take it back

    A pull request merges only when you press Merge, or under auto-merge for safe changes if you switched it on. A pull request is a proposal: closing it undoes it. A task that stays inside the workspace is undone by cancelling its session. In the docs

How the Inbox worksWhat agents can and never do

How is this measured?

What the numbers on this page stand on, from the docs.

Cadence
Each prompt, at most once per engine per country per day. The verbatim answer is kept. Docs
Samples
1x on Starter, 3x on Growth and 3x on Business per reading. Docs
Engines
ChatGPT, Perplexity, Claude and Google AI on every plan; Gemini from Growth. Docs
Markets
Up to 5 countries per workspace, topic or prompt. A country an engine cannot be read from is shown as not measured, never as zero. Docs
What a mention is
An answer that names your brand. A link to your own domain is a citation, counted on the same answers. Docs
Every number
A count with its n beside it. No blended visibility score. Docs
After work ships
Re-read at +7, +14 and +30 days. “No change” is a verdict.
What we do, and what we don't claim
What we doWhat we don't claim
Read each prompt at most once a day per engine and country, and keep every answer word for word.That one day's answer is a trend. A single reading can only be named or not named.
Report mention rate, citation rate and standing separately, each with its n.A single visibility score. There is none.
Show an engine your plan does not read as not measured.That an engine we did not ask gave you zero.

What does it cost?

On every plan · trial on Starter

Reading prompt tracking spends no credits: opening a dashboard, filtering a chart or exporting numbers you already have is free on every plan. Credits pay only for agent work: chats, automation runs and ask calls.

What a piece of work spends scales with how much work it actually does, and Costs in the app shows what the month has cost.

Starter

$129/mo
Prompts
50
Credits a month
3,000
Engines
4
Sampling
1x per reading
Start free

7 days free · card required · $30 of credits (3,000 credits)

See plans and credits

Proof on our own sites

Sites we own · not customers

attensira.comChatGPT + Perplexity

Named in 2 of 1,481answers to the 17 questions that don’t name attensira.com, 2 Oct to 1 Nov 2026 (30 of 31 days sampled). With our own name in the question too: 310 of 2,274.

Firsts
15 firsts. The latest on a question that names no other product:“best AI search visibility tools for agencies”ChatGPT · linked to us ·
The tool’s own wordsChatGPT linked to attensira.com for the first time since we started reading this prompt on 15 September 2026, in an answer to 'best AI search visibility tools for agencies' on 25 October 2026.
Each is a first since we started reading, not proof our work caused it.
Work
8 drafted. Where they stand:
  • 3put live on the site
  • 1live in the workspace
  • 4approved, waiting on us to put live
Re-read, 3 of 3 put live: Nothing we can prove yet
attensira.com's own workspace · read 1 Nov 2026 · n=1,481 answers
modeldiplomat.comChatGPT + Perplexity

Named in 195 of 602answers to the 7 questions that don’t name modeldiplomat.com, 2 Oct to 1 Nov 2026 (30 of 31 days sampled). With our own name in the question too: 595 of 1,276.

Firsts
7 firsts. The latest:“best AI research assistant for debating”Perplexity · linked to us ·
The tool’s own wordsPerplexity linked to modeldiplomat.com for the first time since we started reading this prompt on 1 October 2026, in an answer to 'best AI research assistant for debating' on 16 October 2026.
Each is a first since we started reading, not proof our work caused it.
Work
No receipt yet
modeldiplomat.com's own workspace · read 1 Nov 2026 · n=602 answers

Shown as a baseline: nothing here is offered as a result of our own work.

What does each prompt show?

The same readings, cut the way a buyer question needs. A dash always says why it is a dash.

What does each prompt show?
Mention rateShare of answers that named you, with n, the answers it was counted over. Sampling one answer a day can only ever say 0% or 100%, so read it across a window or a topic.
Citation rateShare of answers that linked a page on your domain. Named without a link and linked without a name are different jobs.
StandingWon, Contested, Lost, Absent or Not measured. Absent is a measurement: we read the answers and nobody you track was named. Not measured is the lack of one.
TopicsA named group of prompts for one buying question. Rates pool every successful read in the group, so a topic carries a signal one prompt cannot.
Filters and exportGroup by topic, intent, standing or market; filter by engine, country and last read. Every cut is in the URL, and the CSV export carries the same standing.
CadenceEach prompt runs at most once per model per country per day, and every answer is kept word for word, so a rate always has readable evidence behind it.

What does prompt tracking not do?

It reports what the engines said, counted honestly. These limits keep a number from meaning more than it does.

  • There is no single visibility score. Mention rate, citation rate and standing are reported separately, each with its sample size.
  • A prompt is read at most once a day per engine and country. Saving it twice does not buy a second reading.
  • An engine your plan does not include is shown as not measured, never as 0%.
  • Discovery never adds a prompt on its own. Approving its Inbox row adds the prompt the same way you would by hand.
  • Removing a prompt keeps its history for your plan's retention window; re-adding the same text starts a fresh prompt.

What does a tracked prompt bring back each day?

The whole answer, every day, on every engine. Perplexity on 4 Oct, to “check if ChatGPT mentions my brand”: it cannot tell, and sends people to articles by Ahrefs, Semrush and SE Ranking. That is the question a tracked prompt answers for you.

Answer · Perplexityasked

Question: check if ChatGPT mentions my brand

Perplexity answered: I can’t determine that from here, but you can check by prompting ChatGPT with your brand name and related queries, or use a brand-mention tracking tool to monitor where and how your brand appears in responses.

Sources cited

  1. 1ahrefs.com
  2. 2brand24.com
  3. 3semrush.com
  4. 4visualping.io
  5. 5brandmentions.com
  6. 6rankprompt.com
  7. +3 more

Not namedAttensira is missing from this Perplexity answer.

Researcher agentResearcher · Stored the answerattensira.com stored Perplexity answer, read over the Attensira MCP · · 10 links

What else do people ask about prompt tracking?

Is this in the trial?
Yes. The Starter trial is 7 days with $30 of credits (3,000 credits): you watch your readings come in and the agent draft real fixes, and you approve what you like. Outreach, automations and top-ups start after the trial.
What does it cost in credits?
Reading prompt tracking spends no credits: opening a dashboard, filtering a chart or exporting numbers you already have is free on every plan. Credits pay only for agent work: chats, automation runs and ask calls. What a piece of work spends scales with how much work it actually does, and Costs in the app shows what the month has cost.
Does anything change on my site without me?
Nothing ships without your yes, unless you let a lane ship on its own. Draft is the default: nothing leaves the workspace until you approve it. In Act, safe changes such as structured data, meta tags, alt text and internal links go ahead on their own, and anything that says something about your business waits for your yes in every mode. A pull request merges only when you press Merge, or under auto-merge for safe changes if you switched it on.
How is this measured?
Each prompt is read at most once per engine per country per day, and every rate carries the number of answers it was counted over. Sampling is 1x on Starter, 3x on Growth and 3x on Business. A change is reported only when it clears a significance test, and work that shipped is re-read at +7, +14 and +30 days.
Which AI engines are prompts read on?
Starter reads ChatGPT, Perplexity, Claude and Google AI every day. Growth and Business add Gemini. Copilot and DeepSeek are not read yet; when they are, they will be named on the models page.
How many prompts should I track?
Enough to group into topics of five or more for each buying question you care about. One prompt on Starter is one reading a day; ten prompts in a topic are ten, which is what makes a change readable.
Should I include my brand name in prompts?
Mostly not. An unbranded prompt is the honest test of whether an engine reaches for you unprompted. Track a few branded ones, such as “what is your brand” or “your brand vs a rival”, on purpose and in their own topic.
Is this an AI Overview tracker?
Google AI Overviews are one of the engines read on every plan, labelled Google AI. Each prompt shows its own cell for it beside ChatGPT, Perplexity and Claude, with the sample size.

See every question you lose, by tomorrow morning.

The moment the trial starts, every prompt you track is read on every engine your plan includes, and you watch the rows fill in. The first report lands once half the reads are in, or after twenty minutes. After that, every prompt is read once a day.

7 days free · card requiredHow the trial works

Start free

7 days free · card requiredHow the trial works