Personal data
Privacy policy
This document describes what the software actually does. Every statement below matches a line of code, not an intention.
Last updated: 21 September 2026.
Data controller
Galaxy SEO, 20 avenue de Bath, 13100 Aix-en-Provence, France. Contact: contact@etalo.ai.
No data protection officer has been appointed: the activity meets none of the three criteria that would require one (public body, large-scale monitoring, large-scale processing of special categories of data).
The public website
On etalo.ai, outside the client area, one thing is collected: what you type into the audit request form.
- What is collected
- Your name, your work email address and the brand to measure — the three required fields. Optionally: company, website, and the free-text message. Nothing else is asked for or inferred.
- Why
- To answer your request, and nothing else. Legal basis: your own approach, which constitutes pre-contractual steps taken at your request (GDPR, article 6.1.b).
- Where it goes
- Two places. An email to contact@etalo.ai, with your address as reply-to. And one line in a log on our server, written before the email is sent: if mail delivery fails, your request is not lost.
- For how long
- Three years after the last exchange, the customary period for commercial prospecting. After that the line is deleted. A request for immediate deletion is honoured within eight days.
- Protection against automated submissions
- A truncated SHA-256 fingerprint of your IP address is kept for one hour, long enough to cap submissions at three per connection. The fingerprint is not reversible and is never matched against any other data.
Audience measurement
We count visits to the public pages. The count is written by our server as the page is sent to you: there is no cookie, no script, no third party, and nothing is stored in your browser. That is also why there is no banner. Here is exactly what it holds.
- What is recorded
- The page viewed and its language, the date, the site you arrived from when your browser says so, the estimated country and city, the device type and the operating system. Nothing else: not the page content, not your movements on the page, not the time spent.
- What is not recorded
- Your IP address. It is used, for the length of the request, to determine the country and city, then discarded: it is written neither to the database nor to any application log. The city is a GeoLite2 estimate, not a location — it often points at your provider's network node.
- How visits are counted
- By a non-reversible fingerprint computed from your address and browser with a salt that changes every day. Two days cannot be joined up: we can say how many visits there were on a Tuesday, never whether the same person came back on Wednesday. That limit is deliberate — it is the price of needing no consent, and it deprives us, by design, of any notion of a unique visitor over a month.
- For how long
- Thirteen months, then automatic deletion. No archive, no backup of those rows beyond that period.
- Robots
- Robot visits — search engines and AI crawlers such as GPTBot or ClaudeBot — are counted separately, from the name they declare and nothing else. They are not geolocated.
- What it is for, and nothing else
- Knowing which pages are read and where they come from, in order to decide what to write next. These figures are not resold, not shared, not matched against a request received through the form, nor against a client account.
The free audit
The "free audit" page asks for two things — a domain and an email address — and it is the only place on the public site that writes to a database.
- What is collected
- The domain you ask to have measured, your email address, and a truncated SHA-256 fingerprint of your IP address. Nothing else: no name, no company, no job title. We also note whether the email address is on the requested domain or not — that is a plain indicator, never a reason to refuse.
- Why
- To send you the result, and to limit new measurements to one per day per connection. Legal basis: your own approach, which counts as pre-contractual steps taken at your request (GDPR, article 6.1.b).
- The measurement itself
- It is kept for seven days and reused if somebody else asks for the same domain within that window. It contains no personal data: it is the public state of a website on a given date, and that date is shown with it. After one year the measurement is deleted.
- Your email address
- Kept for three years after the last exchange, the customary period for commercial prospecting — the same as for the contact form. After that the row is deleted automatically. A request for immediate deletion is honoured within eight days.
- The fingerprint of your IP address
- Kept for 48 hours, the time needed for the daily limit to work, then erased automatically — the request remains, with nothing left to tie it to a connection. The fingerprint is not reversible and is not matched against any other data.
- The result link
- It carries a random token that cannot be guessed. The first opening is timestamped: that is our only proof the email arrived. No tracking pixel, no open tracking — a clicked link says the same thing without tracking anything.
- What we do on the measured site
- Read requests only, like a crawler's: the home page, robots.txt, the sitemap, and a few conventional addresses. Nothing is written, no form is submitted, no account is created.
Access logs of the sites we measure
Inside the client area, a consultant can have Etalo read the server access log of a measured site, to find out which AI crawlers came. An access log carries one IP address per line: here is what becomes of it.
- What is taken from it
- Counts, and nothing else: per day, per crawler and per response code, the number of visits; per crawler, the most-read pages; per day and per channel, the number of visitor arrivals.
- What is never kept
- No IP address, no raw line, no full user-agent, and no URL parameter — the path is cut at the question mark, because an order-confirmation address often carries an identifier, sometimes an email address. The file itself is not kept: a file sent from the interface is read and then disappears with the request.
- The only trace of addresses
- The NUMBER of distinct origins per crawler, computed over blocks of sixty-five thousand addresses (/16). It serves to spot a borrowed crawler name, and it never comes back down to a machine.
- For how long
- The counts are kept for thirteen months, the same period as the site audience measurement, then deleted automatically. The consultant can remove a file and everything it brought at any time.
What the public website does not do
No analytics cookie, no pixel, no analytics script: no Google Analytics, no Matomo, no equivalent. The measurement described above never leaves our server, and no third party receives a copy of it.
No font, script or image loaded from a third party. Typefaces are served by this domain: loading a font from Google would hand your IP address to a third party on every page, which would be hard to defend on a site that otherwise carries no tracker.
No profiling, no automated decision-making, no tracking from one visit to the next.
No resale, no rental, no sharing of your details. No newsletter you did not ask to subscribe to.
The client area
The tool itself, behind a login, processes other data.
- Accounts
- Username, email address, password hash (never the password itself), role and access scope. Kept for the duration of the contract, then twelve months.
- Action log
- Administrative actions are timestamped and tied to an account: creating a client, deleting a campaign, sending credentials. This is what answers "who did what". Kept twelve months.
- Measurement data
- Corpus questions, raw assistant answers, and computed indicators. They concern brands, not individuals — unless a person's name appears inside an AI answer, which is outside our control and which we remove on request.
- Login attempts
- Failed logins are recorded with the username and IP address over a rolling fifteen minutes, to block brute-force attempts.
Recipients and processors
Running the tool means sending corpus questions to third-party assistants. That is the nature of the measurement: you cannot know what an AI answers without asking it.
| Recipient | What is sent | Location |
|---|---|---|
| o2switch | Hosting of the site and database. All data resides there. | France |
| OpenAI | Corpus questions, if ChatGPT is enabled for your campaign. | United States |
| Anthropic | Corpus questions, if Claude is enabled. | United States |
| Corpus questions, if Gemini, AI Overviews or AI Mode are enabled. | United States | |
| Perplexity | Corpus questions, if Perplexity is enabled. | United States |
| SerpApi | Corpus questions, to query AI Overviews, AI Mode and Bing Copilot. | United States |
Corpus questions describe a market — "best site to book a holiday in Greece" — and contain no personal data. Transfers outside the European Union rely on the European Commission's standard contractual clauses, published by each of these providers. The public form is sent to none of these recipients.
Indefinite retention, and why
Raw answers collected from assistants are kept with no time limit, and that is a deliberate design decision.
A visibility measurement only means something when compared over time. Counting rules evolve — a synonym added, a brand mapped to another — and the only way to keep series comparable is to be able to recompute the whole history with the new version. Discarding raw answers would freeze past figures inside a method nobody can audit any more.
These answers concern brands. At the end of a contract, a client may request their complete deletion: it is then final, and the history goes with them.
Your rights
European regulation grants you several. They are exercised by a simple email to contact@etalo.ai, with no formality, and answered within one month.
- Access
- Obtain a copy of the data concerning you.
- Rectification
- Correct inaccurate information.
- Erasure
- Request deletion. It is immediate for a contact request.
- Objection
- Object to a processing operation, in particular any commercial approach.
- Portability
- Receive your data in a machine-readable format.
- Complaint
- Lodge a complaint with the French CNIL (www.cnil.fr) if our answer does not satisfy you. This right does not require contacting us first.
Security
The site is served over HTTPS only. Passwords are stored as hashes, never in clear text. The configuration file, which holds database credentials and API keys, is blocked from web access by a server rule, as is the request log.
Access to the tool is personal: one account per person, with an explicit scope. An account with no scope sees nothing — access is an authorisation, never a default.
In the event of a data breach likely to create a risk to your rights, we notify the CNIL within 72 hours and inform you directly if the risk is high.
Changes
This policy may change along with the software. The update date appears at the top of the page. A change that would broaden the use of your data would be announced to you by email before taking effect.