GEO — your site in AI search answers
ZuBrIQ GEO checks how ready a site is for AI search: whether ChatGPT, Perplexity, Google AI Overviews, Yandex Neuro and Alice will see it, understand it and be able to cite it. In about 20 seconds you get a score out of 100, a list of problems by priority and a fix for each one.
Where to check
Section titled “Where to check”- No sign-in — zubriq.ai/geo: enter the site
address and press Check. Below the result, the Zubr Flash model adds
advice for your site with ready-to-paste fragments: a new title and
description, an llms.txt draft, schema.org markup, robots.txt lines. These
are drafts — review them before publishing. Send the result to a colleague with the
Share button (
zubriq.ai/geo?url=example.com). - In ZuBrIQ chat — ask “check example.com for GEO” or “does ChatGPT see my site”. The model runs the check and writes a fix plan with ready code snippets (the “GEO site audit” skill in the skills catalog).
- Via the API — for agencies and your own systems, see below.
Free check limits on the site: 5 per minute and 20 per day per device and IP. Only public sites on ports 80 and 443 are checked.
What we check
Section titled “What we check”| Check | Weight | What we look at |
|---|---|---|
| HTTPS | 1 | The site opens over HTTPS without errors |
| AI search bots | 3 | robots.txt doesn’t block OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot, Googlebot, Bingbot, YandexBot |
| Training bots | 1 | GPTBot, ClaudeBot, Google-Extended, CCBot — blocking is allowed but lowers the chance of appearing in model answers |
| llms.txt | 1 | A short site “passport” for language models |
| Text in HTML | 3 | The main text is visible without JavaScript — AI crawlers don’t run scripts |
| Title | 1.5 | Present, 10–70 characters |
| Meta description | 1 | Present, 50–170 characters |
| H1 | 1 | Exactly one per page |
| Canonical | 1 | A canonical URL is set |
| schema.org markup | 2 | JSON-LD: Organization, LocalBusiness, Product, Article, FAQPage |
| Open Graph | 1 | og:title and og:image |
| Sitemap | 1 | sitemap.xml is found and returns XML |
| lang attribute | 1 | The page language is set |
The score is the weighted share of passed checks; a warning counts as half. Below 50 — AI search barely sees the site; 50–79 — it sees it, with gaps; 80 and above — the technical side is ready, content decides the rest.
Report by email and help
Section titled “Report by email and help”Under the result on zubriq.ai/geo there is a Send the report by email form. Enter your email to get the score, problems and fixes. Tick I want you to help — the ArtCloud team will write within a business day and offer to adapt the site: text without JavaScript, markup, llms.txt, robots.txt and content. You can also simply reply to the report email.
Key-based API
Section titled “Key-based API”For agencies and anyone checking many sites: the same engine with a ZuBrIQ
key sk-zq-… (the same key as for models, see API). Money is
charged to the key’s organization balance and shows up in Spend as
“GEO checks”.
| What | Price |
|---|---|
| Site check | 0.10 BYN |
| Repeat of the same address within an hour | free (from cache) |
| Weekly monitoring run | 0.10 BYN |
Site-specific advice from Zubr Flash (advice: true) |
0.05 BYN |
| Brand check in AI answers | 0.20 BYN per question |
Limit: 30 checks per minute per organization.
Check a site
Section titled “Check a site”curl https://zubriq.ai/api/v1/geo/check \ -H "Authorization: Bearer sk-zq-..." \ -H "Content-Type: application/json" \ -d '{"url": "example.com", "lang": "en"}'lang is ru or en (the language of titles and fixes). Response:
{ "ok": true, "url": "https://example.com/", "finalUrl": "https://www.example.com/", "score": 62, "fetchedAt": "2026-11-09T16:53:21+00:00", "checks": [ { "id": "ssr_text", "status": "fail", "title": "Text in HTML (SSR)", "detail": "Only 40 words in HTML without JavaScript…", "weight": 3, "fix": "Serve the main text in HTML from the server…" } ], "cached": false, "charged_byn": 0.1, "check_id": 123}status is ok, warn or fail; fix is present only for problems.
With "advice": true the response also carries site-specific advice from
Zubr Flash: advice.summary and advice.items[] with id (the check),
title, advice (what to do) and snippet (a ready-to-paste fragment or an
empty string), plus advice_charged_byn. A repeat within an hour comes from
the cache for free. If the balance is short or the model fails, advice is
null and advice_error is insufficient_funds or advice_failed; failures
are not charged. Advice is stored in the history
(GET /api/v1/geo/checks/{id}).
Errors:
401— no key, or it was revoked;402insufficient_funds— the balance is below the check price;422bad_url— the address doesn’t fit (not http/https, an internal address, a non-standard port);unreachable— the site didn’t respond. We don’t charge for errors;429— the limit is exceeded.
Check history
Section titled “Check history”curl "https://zubriq.ai/api/v1/geo/checks?host=example.com&limit=20" \ -H "Authorization: Bearer sk-zq-..."Returns the organization’s checks, newest first (up to 100): id, source
(api or monitor), url, final_url, host, score, charged_byn,
created_at. All checks of one record — GET /api/v1/geo/checks/{id}.
Do AI assistants mention your brand
Section titled “Do AI assistants mention your brand”The technical audit tells whether AI can read your site. This check tells whether it mentions you when buyers ask about your niche. Give the brand, the domain and up to 5 questions your customers ask. For each question we:
- search the web (Perplexity Search — the same kind of search AI search engines use): is your site in the results, and at what position;
- ask ChatGPT, Gemini and Claude and check whether they name the brand or the domain.
The presence score 0–100 is half the share of model answers that mention you, half the share of questions where the site was found in search. Price — 0.20 BYN per question.
curl https://zubriq.ai/api/v1/geo/brand \ -H "Authorization: Bearer sk-zq-..." \ -H "Content-Type: application/json" \ -d '{ "brand": "Craftsman", "domain": "craftsman.by", "questions": ["Where to order solid wood furniture in Minsk?", "Best carpentry workshops in Belarus"], "lang": "en" }'Response: score, models_share and search_share (0–1); for each question —
search (found, position, url) and answers (model label,
mentioned, a snippet around the mention, answer — the full model answer,
up to 1500 characters). Records are kept: GET /api/v1/geo/brand/{check_id}.
Model answers vary between runs — look at the trend across several checks, not at a single answer.
Monitoring
Section titled “Monitoring”The site is checked once a week. If the score drops by 5 or more, or a new critical problem appears, you get an email: what broke and how to fix it.
# enablecurl https://zubriq.ai/api/v1/geo/monitors \ -H "Authorization: Bearer sk-zq-..." \ -H "Content-Type: application/json" \ -d '{"url": "example.com", "email": "seo@agency.com"}'
# listcurl https://zubriq.ai/api/v1/geo/monitors -H "Authorization: Bearer sk-zq-..."
# disable (history stays)curl -X DELETE https://zubriq.ai/api/v1/geo/monitors/42 \ -H "Authorization: Bearer sk-zq-..."emailis optional: without it the email goes to whoever issued the key.- The first check runs the next morning (07:20 Minsk time), then every 7 days.
- Up to 50 monitors per organization.
- Not enough money — the monitor skips the day (
last_error: insufficient_funds) and tries again the next morning.