Help centre
The questions that come up
Thirty-nine answers, grouped by theme. If yours is missing, write in: it will probably end up on this page.
The measurement
What exactly does Etalo measure?
Your brand's presence inside AI assistants' answers. Concretely: a corpus of questions representative of your market is put to ChatGPT, Claude, Perplexity, Google AI Overviews, Google AI Mode and Bing Copilot, several times over. Etalo then counts how many answers name your brand, in what position, and which other brands appear alongside — or instead of — you.
How is this different from Google rank tracking?
Rank tracking measures where a page sits in a list of links. Etalo measures whether your brand is named in a written answer. The two are not correlated: a page can rank first on Google and be absent from an assistant's answer, because the assistant read three pages and kept only two brands.
What is share of voice?
The number of times your brand is cited, relative to all brand citations in the same answers. If 100 answers contain 56 brand citations of which 18 are yours, your share of voice is 18/56, or 32%. It is read alongside the presence rate, which answers a different question: in what proportion of answers do you appear at all.
What does "domain citation" mean?
When the AI talks about your brand, does it rely on your site, or only on what third parties say? A low figure here while your share of voice is fine signals a technical problem — your site is poorly read by generative crawlers — rather than an awareness problem.
Why is each question asked several times?
Because an AI never answers the same way twice. A single answer supports no inference. With three passes, you classify qualitatively: cited on all three, on one or two, or never. With more passes, a rate is published with a 95% confidence interval.
Is a two-point gap between two months significant?
Probably not. Every rate is published with its margin of error, and a gap smaller than that margin is not a movement: it is noise. The report shows the interval precisely so that nobody comments on noise in a meeting.
What does the accuracy rate measure?
The share of claims about your brand that survive verification. Every claim is checked against a truth sheet you approve — prices, scope, locations, executives — and inaccuracies come out ranked by severity, with the answer they came from. Unverifiable claims are excluded from the rate rather than counted as correct: including them would inflate the figure.
What are tone and perceived positioning?
Tone says whether the passage citing you is favourable, neutral or unfavourable. Perceived positioning says where the assistant places you: the market reference, an alternative, one option among others. Two brands can hold the same share of voice and be presented in completely different ways.
What is the difference between being named and being linked?
A bare mention names your brand without a link: it builds awareness and brings no visit. A linked mention opens a click path, and shows up in your logs. There is a third, more frustrating case: your site is used as a source without the brand being named — the content carries authority, the company does not exist in the text. All three are fixed differently, so they are counted separately.
The method
What is a corpus, and why is it frozen?
The corpus is the list of questions put to the assistants. It is frozen before the first campaign and does not change afterwards. That is what makes two campaigns comparable: a corpus regenerated every time would measure different questions and produce nothing but noise. Adding questions is possible, but it opens a new corpus version, and the series restart from there.
Who writes the questions?
They are built from a 200-template sector pack, then adapted to your buyers' real vocabulary during a one-hour interview. You review and approve them before the freeze. It is the only step that takes time on your side, and it determines the quality of everything that follows.
Do the questions cover the whole buying journey?
Yes, in four stages: discovery ("how to choose"), exploration ("what solutions for"), comparison ("X or Y"), selection ("best site for"). A competitor may dominate comparison while you hold discovery — the gap is read stage by stage.
What is a "derived query"?
An assistant never searches the question it was asked. It breaks it into several queries, reads the pages those return, then writes. Etalo captures the queries actually issued and the domains returned. A frequent query where your domain never appears is a page to write — and it is written for that phrasing, not for the original question.
What do "memory mode" and "augmented mode" mean?
In memory mode, the assistant answers without searching the web: it draws on what its training retained. In augmented mode, it runs searches before writing. The two give very different results, and both count: many users never turn web search on.
Do you query assistants from a user account?
No. Calls go through the vendors' programming interfaces, with no conversation history and no personalisation. That is deliberate: a measurement depending on one account's habits would be comparable to nothing.
The report and deliverables
What does a monthly report contain?
Your share of voice and its movement, the competitive map by assistant and by stage, lost questions with the incumbent to displace, derived queries and the pages read instead of yours, factual errors AI makes about your brand, and a correction plan prioritised by impact and effort.
Can I check a figure that looks wrong?
Yes, and it is designed for that. Any percentage drills down to the exact sentence that produced it, with the assistant, the mode, the model and the timestamp. If a figure surprises you, you can reach the raw answer in three clicks.
In what format can I get the data?
Everything exports to CSV from the interface, at any time and at no charge: the corpus, raw answers, indicators, cited sources, correction plan. The full report also downloads as a laid-out, paginated PDF: cover, summary, campaign figures, technical diagnosis and correction plan in a single document.
Am I told when something changes?
Yes, by email, the same day, in both directions. Five situations raise an alert: an AI crawler that can no longer read your site, your score dropping, an axis falling below half its points, a competitor overtaking you, your citation rate falling. The same five measurements, read the other way, report progress: a crawler allowed again, an axis back above half, a place gained. It all fits in a single message, and the same finding does not repeat for a fortnight: an alert people end up filtering is no use to anyone.
Can my colleagues have access?
Yes. Read-only accounts are free and take a minute to create, scoped by brand and country. Your agency can have one too. We prefer that to sharing a single login, which makes the action log useless.
What happens if an assistant fails during collection?
The engine concerned is set aside and flagged in the report, rather than silently omitted. An interrupted collection resumes exactly where it stopped, without replaying calls already made.
What are the AI Score and the AI Benchmark?
The AI Score is a technical audit of your site: eighteen checks across four axes — AI crawler access, discovery, content readability, ability to act — each with its evidence and its recommendation. A check that could not be run leaves the scale instead of counting zero. The Benchmark replays the same audit on the competitors assistants actually cite, on the same day, and compares only the checks evaluated everywhere.
Automating and exporting
Do campaigns have to be launched by hand?
By default yes, and deliberately so: paid API calls are not triggered without a human deciding. But each client can be given a start date and a frequency — daily, weekly, fortnightly or monthly. The next campaign then creates itself, arms itself and collects, on the frozen corpus and with the same assistants as the previous one.
Can I choose which assistants are queried?
Yes, campaign by campaign. Ten surfaces are available — ChatGPT, Claude, Gemini, Perplexity, Mistral Le Chat, Grok, DeepSeek, Google AI Overviews, Google AI Mode and Bing Copilot — and you decide which enter the measurement. The screen shows the number of API calls before you confirm, because that number is what drives the cost.
Can an assistant be missing from the list?
Yes, and the screen says so rather than hiding it: an engine whose API key is not configured appears as excluded, with the reason. That is what prevents a campaign measuring five surfaces while the report announces six.
How do I plug the figures into my dashboard?
Thirteen CSV feeds are available from the "Data feeds" screen, from the campaign summary down to which page was cited, on which question, by which assistant. Each feed has its own URL, readable without signing in, ready to paste into Looker Studio, a spreadsheet or a warehouse. The screen also gives the ready-made IMPORTDATA formula.
Do those URLs expose my data to anyone?
They carry aggregates only: no answer text, no claim, no verdict passes through them. The address contains a long token, specific to your brand and revocable — regenerating it cuts the old ones immediately. That is the accepted trade-off of a feed readable without a login, and why the sensitive detail is absent from it.
Is the report commented, or only numbers?
A campaign analysis is written from that campaign's figures, in your language: what moved since the previous one, what explains it, and what to fix first. It is reviewed before it goes out, and it never states a figure that was not measured.
Your data
Is my data used to train a model?
No. Neither by us nor for us. Corpus questions pass through AI vendors' interfaces to obtain an answer — that is the very principle of the measurement — but they describe a market and contain no personal data.
How long do you keep the collected answers?
Indefinitely, and that is a deliberate choice. Counting rules evolve, and the only way to keep series comparable is to be able to recompute the whole history with the new version. At the end of a contract you may request their complete deletion: it is then final.
Is my data mixed with other clients'?
No. No aggregation, no sector statistics published from your data, no cross comparison. Each client sees their own scope and nothing else.
Does the public website set trackers?
None. No pixel, no social button, no third-party script. Typefaces are served by this domain, so that your IP address is not handed to a third party. The site's only cookie keeps your session once you are logged into the tool. Details in the cookie policy.
Do you count visits on this site?
Yes, and without cookies: our server writes one row as it sends you the page. It holds the page read, its language, the date, the site you came from, the estimated country and city, the device and the operating system. Your IP address is used to determine the country and then discarded, and the fingerprint used to count visits changes salt every day — so two days cannot be joined up. Rows are deleted after thirteen months. That is what spares you a banner: nothing is written to or read from your browser.
Getting started
How long before the first figures?
About two weeks. A one-hour interview to frame the scope, a few days to build the corpus and have it approved, then the first campaign. You then receive the report and the correction plan.
What do you need to know about my brand to start?
Your ranges, your markets and your real competitors — the ones your customers mention, not the ones in the sector ranking. Plus the variants of your brand name, including common misspellings: a misspelled brand in an answer still counts.
Do I need accounts with OpenAI, Anthropic or Google?
No. API costs are included in the engagement. You have no account to create, no key to manage, no vendor invoice to watch.
How much does it cost?
By quote, because a question set is built for a specific market. Four variables determine the price: set size, number of assistants queried, number of markets, and frequency. The Pricing page details them. The AI Score and the Benchmark rest on plain HTTP requests, whereas a campaign commits thousands of metered API calls: that is what separates the two, and why the demonstration runs on the first.
Do you guarantee my brand will be cited more?
No, and nobody can: no one controls what these models answer. What is guaranteed is knowing where you stand, against whom, on which precise questions — and receiving an action plan where every line rests on an observation rather than a hunch.
I lost my password, what now?
The sign-in page carries a "forgotten password" link: you receive a single-use link valid for one hour. For safety the page answers the same whether the address exists or not — otherwise the form would tell anyone who is a customer. If nothing arrives, write to us.
Your question is not here?
Write it in. Useful answers end up on this page.