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`.
Attensira Logo
Attensira
Attensira for Brands

Be the sentence
the model writes.

Attensira measures how often ChatGPT, Perplexity, Google AI, Claude and Gemini name your brand and link your domain — with the sample size attached — then writes the page fixes and opens a pull request for you to review.

Cancel anytimeNothing merges without your approval

A model answer is not a results page

There is no position ten to climb from. The model either names your brand in the sentence it writes, or it does not — and it may answer the same question differently the next time it is asked. That changes what can honestly be measured, and what a fix looks like.

Google
OpenAI

There is no rank, only a rate

Attensira reports how often an answer named you and how often it linked your domain, each with the number of successful runs behind it. No blended visibility score, because there is nothing to blend.

🤖
AI crawler fetches
GPTBot fetched /docs
PerplexityBot fetched /pricing
ClaudeBot fetched /compare
Google-Extended fetched /docs

Crawlers do not run your JavaScript

Which is why a browser-beacon analytics tool shows zero AI crawler traffic — it is describing itself. Attensira counts these fetches at the edge, and keeps them separate from citations.

Mention raten = 3
33%
Change vs. last week: not provable
Example

Small samples get said out loud

33% at n = 3 is one hit in three draws, so Attensira prints the n. Every delta goes through a two-proportion z-test at 95%; below that it reports no provable change rather than a trend line.

Worked example — a fictional company

What one reading looks like

Attensira has no public customers yet, so this is an illustration, not a result. Northwind CRM tracks the prompt below. Each model answers it once per day per country, and the answer text is stored exactly as returned.

Two counts per answer, never blended

A mention is the brand named in the text. A citation is a URL the model attached that points at the brand's own domain. An answer can do either, both, or neither, so the two rates are reported separately — each with the number of successful runs behind it.

n
Successful runs, shown with every rate
null
Not measured — never drawn as a zero
95%
Bar a change must clear to be reported
OpenAI
ChatGPT

“What's the best CRM for a five-person sales team?”

Named Northwind CRM in the text, no link attached — counted as a mention, not a citation.

Perplexity
Perplexity

Attached a link to northwindcrm.com/pricing — a citation, and one row in the Pages view.

What you get, screen by screen

See which brands the models name instead of you

Add a competitor by domain and Attensira scores their mention and citation rates over the answers it has already collected — same window, same denominator, no extra runs and no prompt slots spent. The rates are independent, so they can sum past 100%: one answer can name four brands.

See how tracking works
Mention rate · this window · n = 42 runs
Northwind CRM (you)
1638%+9 pts
competitor-a.com
1228%not provable
competitor-b.com
819%not provable
competitor-c.com
49%-11 pts
Independent rates over the same 42 answers — not shares of a fixed total, so they can sum past 100%.
5
AI answer surfaces read: ChatGPT, Perplexity, Google AI, Claude, Gemini
n
Sample size published beside every rate
95%
Significance bar before a change is reported
0
Pull requests merged without you
Built for Brands

Measurement you can check, work you can review

Four things Attensira does that a reporting dashboard stops short of.

Read the answer, not a score

Every successful run is stored verbatim, so you can read the paragraph that named a competitor instead of you. Detection runs over that stored text, which means a rule change can be re-run across your history rather than starting a new chart.

See how tracking works

See which of your URLs get cited

The Pages view lists your own cited URLs with their citations in the window, their share of your domain's citations, and their average position inside the model's citation list. That is the shortlist the agent works from.

See page-level citations

Get the edit as a pull request

The agent writes the change and opens a PR on a fresh branch, labelled `attensira`, whose body names the page, the finding, and why the edit answers it. It then tracks whether you merged or closed it and what CI said.

See how the agent writes

Know which domains the models trust

Every cited domain is bucketed as your own, a competitor's, an AI platform, a reference site, or a third party. Third party is the narrowest bucket and the only one where a citation is genuinely winnable — that is where the work goes.

Explore prompt data

See what the models
answer about you today.

Add the buyer questions you want to win, connect the repository your site lives in, and read the first pull request the agent opens.

Cancel anytime
FAQ

Frequently Asked Questions

Want to see a real pull request first?
Book a walkthrough and we will run your category live.
Book A Demo

Two things per answer, counted independently. A mention is your brand named in the text. A citation is the model attaching a URL that points at your own domain. Attensira runs your tracked prompts against ChatGPT, Perplexity, Google AI, Claude and Gemini on a schedule, stores the verbatim answer, and reports mention rate and citation rate with the number of successful runs behind each.

No. There is no blended visibility score and no sentiment metric, and nothing divides a fixed total between brands. Competitor rates are calculated over the same answers and the same denominator as yours, which means they are independent rates and can sum past 100% — one answer can name four brands.

There is no rank. A model either names you in its answer or it does not, so the reading is a rate over sampled answers rather than a position. The only position Attensira reports is average position, which is how far down a model's own citation list a given source domain tends to sit — per domain, not a leaderboard of brands.

By sampling and by refusing to over-read. The same prompt goes to the same model once per reading on Starter and three times on Growth and Business, and every reported change is put through a two-proportion z-test at 95%. Below that bar Attensira reports no provable change rather than a number.

That is the point of it. It writes the page edits its own findings call for and opens a pull request against the repository the page lives in, through a GitHub App you install into your own org. It never merges, never pushes to your default branch, and never touches CI configuration.