In brief: geoscan.seo7.es is the studio's free tool. You enter a website address and in about 10 seconds you see whether ChatGPT, Perplexity, Gemini and Copilot can find, read and cite your page. The check runs 172 tests in 8 groups and answers with two independent numbers instead of a single overall score.
Client: SEO7's own project · Site: https://geoscan.seo7.es · Category: GEO / AI tool
What matters most in this project?
- 172 checks in 8 groups: seven groups about the page itself, the eighth about the brand's presence outside the site.
- Two numbers instead of one: whether the page is ready to be cited, and whether the model knows your brand before it opens the page.
- What does not apply does not count against you: a recipe schema on a lawyer's site stays out of the score, so an honest hundred is reachable.
- Reality axis: 6 real queries to AI about your niche, plus the competitor domains the model names instead of you.
- Result on screen for free: the PDF report arrives by email in Spanish, Russian or English.
- The enquiry lands in the studio CRM together with the site address and the check results.
What is the project built with?
- Frontend:
- a single page in plain HTML, CSS and Vanilla JavaScript, no frameworks, in three languages.
- Pipeline:
- self-hosted n8n in Docker, with the scan split into visible steps: fetch, parse, access, citability, markup, content, technical, external signals and Reality.
- Language model:
- DeepSeek handles the Reality axis.
- PDF:
- a separate Node and pdfmake service with Cyrillic support.
- Infrastructure:
- nginx with TLS on the same VPS that runs the studio backend.
- One engine:
- the same calculation runs in the analyser on seo7.es and in the daily check of every page on the studio site.
Each pipeline step can be opened and run on its own. When the whole calculation hides in one large block of code, an error only shows up in the final score; the split pipeline shows the intermediate result after every step.
How did it start?
People stopped googling and started asking. "Recommend a dog groomer nearby" or "a good garage in Valencia" gets a ready answer with company names, and your business is either in that answer or it is not.
A site owner cannot check this on their own. Regular SEO services measure search positions, not whether ChatGPT reads the site or whether the model has enough data to recommend you.
Why did the engine need a new version?
The first version gave our own site 96 out of 100, and yet the model did not name it in any of six queries. A client site showed the same picture: 97 out of 100 and zero citations. The score measured technical hygiene that a decent site already has in order, and said nothing about what really matters.
Research shows brand mentions affect citability three times more than links, and the model takes 85% of mentions from third-party sites rather than from the site itself. So version 1.2 answers with two numbers: whether the page is ready, and whether the model knows the brand. A single score would hide the second behind the first.
What was built?
🌐 How is the interface built?
- A check from a single address, with no sign-up and nothing to install
- Three languages, results in the user's language
- A cookie notice, localised links and a dedicated mobile layout
🧠 How are the two numbers calculated?
- Citability: content is multiplied by bot access, not added to it
- Brand recognition: with fewer than three independent external signals the tool returns an empty value, not a low number
- 8 groups weighted by impact, a score from 0 to 100 and a grade from A to F
- Checks that do not apply drop out of the denominator
🤖 What does the Reality axis show?
- 6 real queries to the language model about the site's niche
- Whether the model named your brand or your competitors
- The exact domains that are taking your place
⚡ How is the full mode different?
- It adds loading speed, external mentions and a live query to four models
- It takes up to two minutes instead of ten seconds
📄 What is in the report?
- The score by group and all 172 checks with result and description
- The Reality axis and priority fixes, ready for a developer
- On screen straight away, the full report as a PDF by email
📚 Where is the method described?
- The "Methodology" section on geoscan.seo7.es in three languages
- A working paper on the method on Academia
- A methodology repository on GitHub under a CC BY 4.0 licence
What is the result?
| Metric | Value |
|---|---|
| Checks | 172 in 8 groups |
| Quick check | about 10 seconds |
| Full mode | up to 2 minutes |
| AI queries on the Reality axis | 6 |
| Numbers in the answer | 2: citability and brand recognition |
| Interface languages | español, русский, English |
| Cost to the user | €0 |
| Where the enquiry goes | straight into the studio CRM |
| Openness of the method | methodology on the site, paper on Academia, repository on GitHub (CC BY 4.0) |