AI citation intelligence
AI already recommends brands. Measure whether it names yours.
We query the models behind ChatGPT, Claude, Gemini and Perplexity through their official APIs, with web search enabled. You get how many responses mention you, your share against competitors, and which of your pages get cited.
The four engines we measure, through their official API:
- ChatGPTAPI with web search enabled
- ClaudeAPI with web search enabled
- GeminiAPI with web search enabled
- PerplexityAPI with web search enabled
Green means web search on; grey, off. We never say «we ask ChatGPT», because it is not the same.
Illustration of the dashboard with example data. The same figures appear as text, with their denominators, in the example report section of this page.
What Citenza does
Track your citations
How many answers you appear in, in what words, and which of your pages get linked. By engine, language and country.
See the reportBenchmark your competitors
Share of voice, presence rate and position against the brands you choose to track, with 30 and 90 day series.
See the comparisonUnderstand how it is calculated
Model, parameters, country, repetitions and formula version. Published in full, including what the product cannot do.
Read the methodology
Free scanner
Your score in about 60 seconds, no signup.
This is not identical to asking in the ChatGPT app, and we explain why, with the exact parameters of every query. How we measure it
How it works
Three steps, and the first one is free.
From your brand name to a report with denominators, with no sales call in between.
Add your brand
Name, aliases, domains and up to ten competitors. Nothing else: the system proposes the prompts and you approve them.
We run the prompts
Thousands of real answers through each engine's official API, with web search in the state we declare, storing the full response.
You get the report
Every figure with its denominator, its confidence interval and the fidelity of the engine that produced it. Exportable and auditable.
Definitions
Transparent metrics, not vanity numbers
Three figures, and all three carry the count they come from. A percentage without a denominator is not a measurement: it is an opinion with decimals.
Share of Voice
27.3%
Of every appearance by the entities you track, what share is yours. Each response counts at most once per entity.
164 of 600 mentions
Presence rate
13.7%
In what share of ALL responses in the slice you appear, including those that mention nobody. This is absolute visibility.
164 of 1,200 responses
Visibility score
44/100
- Invisible0 to 19
- Emerging20 to 39
- Present40 to 59
- Strong60 to 79
- Leader80 to 100
The example scores «Present» despite a 27.3 % share of voice, because presence carries 40 of the 100 points. That is the figure the free scanner returns for your brand.
How the presence rate is calculated
How the presence rate is calculated
Every answer in the period counts, not only the ones where somebody shows up. Answers that mention no brand at all are answers too, and leaving them out is what turns a low figure into a pretty one.
- Answered with your brand cited9 answers
- Answered without your brand7 answers
- No usable answer4 answers
Example with 20 answers
- Answered with your brand cited
- Answered with your brand cited
- Answered with your brand cited
- Answered with your brand cited
- Answered with your brand cited
- Answered with your brand cited
- Answered with your brand cited
- Answered with your brand cited
- Answered with your brand cited
- Answered without your brand
- Answered without your brand
- Answered without your brand
- Answered without your brand
- Answered without your brand
- Answered without your brand
- Answered without your brand
- No usable answer
- No usable answer
- No usable answer
- No usable answer
Presence rate
45.0%
= 9 of 20 answers
- Presence rate
presence_rate = brand_hits / total_responses - Your mentions across ALL responses in the period, including the ones that name no brand at all. This is absolute visibility, and it is the figure that stings.
- Share of Voice
share_of_voice = brand_hits / all_tracked_hits - Your mentions against the mentions of every tracked brand. Each response contributes at most one mention per brand, even if the text repeats your name eight times.
- Citation rate
citation_rate = cited_responses / citable_responses - Responses that link to a domain of yours, against the responses from engines that can cite at all. Engines that cannot cite stay out of the denominator: averaging them in would produce a meaningless number.
Model, parameters, country, repetitions and formula version. Published in full, not summarised.
Example data
This is what you get to see
One example brand against four competitors, over 1,200 responses across 30 days. The figures are made up. The way they are presented is exactly what the product does.
Share of voice against competitors
- Competitor A34.7%208 mentions
- Your brandYou27.3%164 mentions
- Competitor B19.7%118 mentions
- Competitor C12.0%72 mentions
- Competitor D6.3%38 mentions
Denominator: 600 mentions across the five tracked brands, spread over 1,200 responses in the period.
The two figures we never split
- Share of Voice
- 27.3%
- 164 of 600 mentions
- Presence rate
- 13.7%
- 164 of 1,200 responses
Each square is 1% of the period responses. 14 filled for a 13.7% presence rate: the fraction is rounded up, the figure above is not.
Share of voice tells you how the pie is split. Presence rate tells you whether there is a pie. In this example, 1,036 responses mention nobody at all.
Eight weeks, compared
Who it is for
Built for teams who need proof
The four profiles already using this data, and what they use it for.
SEO teams
Explain the drop in organic traffic with data, not a hypothesis. And find the queries being resolved inside a chat with no click at all.
Communications
Measure whether the published narrative reaches model answers, and with what sentiment. It is the effect of earned media, counted in answers.
Agencies
White labelled reports, client sub-accounts and a published methodology that saves you the first half hour of every meeting.
Brand and product
Know how a model describes you when nobody asks about you, and which competitors it pairs you with when comparing.
We have no clients to show yet. This is what you can verify
The product has just been born. You will find no bought logos and no anonymous testimonials here: six things you can verify yourself before paying, and the two most uncomfortable sentences we say about ourselves.
- Full methodology published, not summarised
- Every figure shows its denominator
- Fidelity declared engine by engine
- Data on our own server in Spain
- Auditable export and report
- Formula version stored with every report
This is not identical to asking inside the ChatGPT app, and we explain exactly how it differs.
Methodology, fidelity sectionA 100 % share of voice can be 3 mentions in 300 answers. That is why the two figures always travel together.
Methodology, metric definitions
The prices are right here, not behind a sales call
14-day trial, no card required. The trial budget is the lowest in the system: we limit spend, not features.
Starter
€49per month, excluding VAT
- 1 brand
- 50 active prompts
- 2 engines
- Weekly sweep
Pro
Recommended€149per month, excluding VAT
- 3 brands
- 200 active prompts
- 4 engines
- Weekly sweep plus daily core
- Email alerts with statistical testing
- Public read-only API
Agency
€499per month, excluding VAT
- 10 brands
- 1,000 active prompts
- Unlimited client sub-accounts
- White-labelled reports
We sell only to businesses and professionals. At checkout we ask for your VAT number; for EU businesses with a valid VAT number, reverse charge applies and no VAT is added.
Invoice amounts are calculated by Stripe according to your country and tax situation. We never recalculate an invoice.
Objections
What every agency asks on the first call
Do you query ChatGPT?
Not exactly, and the difference matters. We call the official API of the model that powers ChatGPT, with web search enabled and no system prompt of our own beyond a neutral one we publish. The consumer product adds memory, personalisation and experiments we cannot reproduce. That is why the value lies in the comparison over time against your competitors, which is robust to that difference.
Can I trust a figure if the answers change every time?
The answers are not deterministic, and we do not hide that. That is why every metric is shown with its n, series carry Wilson confidence intervals, and the smallest unit of comparison is the seven-day window. We never present the difference between two consecutive runs as a trend.
Where is my data?
On our own server in Spain, in the European Union. Prompts are sent to the AI providers being measured, which operate in the United States: that is inherent to the service and is detailed in our list of subprocessors. We do not use your data or the responses to train anything.
Why is the full sweep weekly and not daily?
Because a daily chart of 200 prompts would mostly be expensive sampling noise. The weekly signal is the one with statistical meaning, and the daily sweep of a core of prompts you choose is what feeds the alerts. It is a better product, and it costs less.
Does the trial ask for a card?
No. It is 14 days with the Pro plan's features and the lowest usage budget in the system. We limit the money, not the features.
Start by knowing where you stand
The scanner runs 10 real responses and returns your score in about 60 seconds, with the breakdown in front of you. No card, no signup to see the number.