Skip to content
SEO Records
Choisir la langue : English

GEO SEO Records

Your site open to AI assistants. And the measure of what they bring in.

GEO SEO Records opens the door to the bots that read on behalf of a human, lays out the map and the answer, declares who is speaking — then counts who comes in, and what it pays. Counting is the promise that holds. It is also the one that produced a number.

What is measuredWhat you get →

Five pieces, in the order in which they count.

Working on the answer before the door means writing behind a closed door. The first four make the channel possible; the fifth is the only one that tells you whether it exists.

01

The door

Can the crawlers that read on behalf of a human get in? We read the written rule (robots.txt) AND we knock, using each AI assistant's agent, alongside a neutral witness. A firewall can say no while the rule says yes: we knock, and we see.

02

The map

Does your llms.txt present only what can be verified? Every link in it is knocked on, and every promise is sorted: the one a machine can check, and the one it falls to you to settle — a fact or an argument.

03

The answer

Do your pages state questions and their answers, and do they answer right below the heading rather than after four thousand characters? An AI assistant quotes a short, attributable answer before a well-ranked page.

04

The citable, the entity, the structure

Figures that are dated and sourced rather than bare; a declared publisher, with the same name everywhere and third-party profiles that genuinely point to it; blocks of forty to sixty words that stay true once torn from their page.

05

The channel — and what it brings in

The only piece that tells you whether the channel exists. Placed in the site or at the edge of the network, it counts who comes to read, separates the human visit, the live read and the harvest — each counted separately — and links the pages read to the euros collected.

The records — reading of 09/05/2026

Five records, one order each.

A record can be replayed: every figure comes from a command, and the command is written right beside it. The day a figure failed, its line would leave this page.

0 / 0false positive / miss

Across 15 pages from 13 real sites, 120 codes judged

Pages from real sites frozen on 09/05/2026 with their provenance, in French, English and German: Shopify, WooCommerce and PrestaShop stores, press, government, encyclopedia, documentation. Ground truth declared on evidence, page by page, before being written: every code issued on a real page is earned, and the command replays it.

npm run conditions-reelles

5certified channels · 816 ranges

A live read that truly comes from the publisher

"ChatGPT-User" is a word in a header, and a word can be written. OpenAI, Perplexity, Apple, Microsoft and DuckDuckGo publish their crawlers' addresses: the address of every read is checked against those ranges as it passes, only the verdict is recorded, and the address is forgotten immediately. A certified read comes from its publisher's own network, and the log proves it line by line.

npm run plages

12 / 12in 403 · control 200

The real door, with a control

Reading taken on 09/05/2026 on nytimes.com: robots.txt blocks seven live-read crawlers, and the firewall returns 403 to the twelve AI assistant agents while the neutral agent gets through. The blocking targets them, and the measurement proves it: two layers, two figures, two causes — read in a single command.

node outils/analyser.mjs https://www.nytimes.com

0 €of API per month

Zero euros of API: the site's logs are enough

Visibility platforms charge from 29 to 489 dollars a month because they pay the models to query them. We read the site's logs: everything is already there. One million log lines are re-read in 601 ms.

npm run records

44cases, two languages, 0 divergence

The same measurement in Node, at the network edge, and in PHP

The PHP wrapper tables are generated from the Node version, and parity cases are computed by it then replayed in PHP. Tested on a WordPress installed in a subdirectory and on a PrestaShop 8. A table duplicated across two languages would be a truth held twice.

npm run php:tests

Firsts

Three mechanisms we alone deliver.

Each one is in the product today, and replays on your side with your own log. The day one of them disappeared, its line would leave this page.

01

Reads certified by address, written into the site

Every read by a declared crawler is checked against the IP ranges published by its operator at the moment it comes through, and the verdict is written line by line in the site's own log — with zero cookies and zero trackers — then joined to the pages that pay. The only AI assistant channel log that certifies what it counts.

02

Demand read, rather than guessed

When an AI assistant comes to read a page, a human has just asked a question on that topic, right now. Live reads, per page and per day, tell you what people come looking for on your site: real questions, read in your own log, at zero euros of API.

03

The causal trial

Equipping some pages, leaving others, and comparing certified reads before and after, with a permutation test on a published seed. The only assumption: the labels could have fallen differently. The result replays for anyone holding the log. The experiment requires hands on the site: we have them.

What you get.

  • The door open to crawlers that read on behalf of a human, verified by knocking; the map and the response in place; the publisher declared, with the same name everywhere.
  • The exact count of who comes in: visits, certified live reads, harvests — each in its place — and the pages that pay, joined to the euros collected.
  • Demand read in your log, topic by topic, and the causal trial to prove what a piece of equipment brings in. Everything replays, everything is dated.
  • Optionally, with your key and at your provider: the probe, which tells you whether a model cites you from memory, question by question, and who it cites in your place.

Items from the September 5, 2026 reading: OpenAI's published ranges, from Perplexity, from DuckDuckGo, and robots.txt knocked on.

The market's entry price, for the whole log. Setup included.

You pay once, for one year, and the price shown is the price. The WordPress plugin or the PrestaShop module is sent by e-mail the moment payment is made, checksum-verified from our signed channel; your log's address and its secret appear in your admin panel as soon as it's activated. Another platform: we install the measurement at the network edge within 24 hours. The price includes installation, updates and support, all year round.

One site

490 €for one year

Paid once: the year runs day for day, i.e. €41 per month. One domain, measurement set up within 24 h, certified log, request read, causal test, updates and support included.

Take a year

Three sites

Recommended

1 190 €for one year

Paid once, for one year. Three domains, platforms of your choosing, i.e. ~€397 per site. Each one set up, measured and supported separately.

Take a year

Agency

2 990 €for one year, up to 10 sites

Paid once, for one year: less than a month of agency fees, for ten sites. Ten fully supported domains, your clients' or your own: each one gets its setup, its log and its support, and we stand behind them.

Take a year

The log stays with you and keeps four things only: the channel, the type, the path, the time to the minute. You read it whenever you want, with your secret; a visitor's address is used once, for certification, then forgotten.