Principles
- Anything that can be checked in code (robots.txt, llms.txt, schema, author and dates, consent platforms, Core Web Vitals) is checked in code. The model does not re-score these facts, so the same URL gets the same measured part of the score.
- The language model (Claude via OpenRouter) only judges content criteria — e.g. whether a paragraph stands on its own or a section opens with a direct answer — and writes suggested fixes. This part may vary slightly between runs.
- Suggested rewrites use only facts from your page. Missing data is marked [add data] — the tools do not invent numbers, sources or quotes.
- No tool measures actual citations in ChatGPT, Perplexity or Google. Scores describe how ready a page is; they do not guarantee rankings or mentions.
- Legal tools (GDPR, EU regulations, AI Act) are technical reviews and do not replace legal advice.
Tool by tool
| Tool | Measured in code | Judged by an AI model | Limits |
|---|---|---|---|
| SEO audit | 19 checks: indexability, HTTPS, viewport, canonical, title, meta description, H1/H2, internal links, alt text, schema, Open Graph. The score is computed from them. | Concrete fixes and suggested copy for failed checks. | The score covers the home page; up to 4 sitemap subpages are checked for recurring issues. No JavaScript rendering. |
| GEO audit | AI search access in robots.txt, llms.txt, schema types, FAQ, author, dates, freshness (6 of 10 factors). | Sentence citability, concrete data, direct answers, entity clarity (4 factors). | Does not measure actual citations in AI answers. |
| AEO review | Schema, structure (lists, tables, questions), authorship, bot access. | Conversational match, answer conciseness, entity clarity, direct-answer opportunity. | Uses a sample of the first ~3,000 characters of text. |
| AI crawler access | robots.txt rules for 18 bots per RFC 9309 (search indexes, live fetchers, training) and llms.txt presence. | — | A CDN or firewall can block bots even when robots.txt allows them. |
| llms.txt validator | Structure per the llmstxt.org proposal: H1, summary, sections, links, descriptions, size. | — | llms.txt is a community proposal; major AI providers have not confirmed using it. |
| Citability by paragraph | Paragraph extraction and selection of the five longest. | Five editorial criteria per paragraph and a rewrite without invented facts. | An editorial judgement, not a measured citation probability. |
| Text citability | Sentence splitting and counting (code). | A label per sentence (citable / partly / not) and overall criteria. | Long texts are sampled evenly up to 150 sentences. |
| Brand mentions in AI answers | Counting mentions and their position in 16 answers (code). | Writing 4 questions, answers from 4 models, tone of mentions and recommendations. | Models are asked without web search — the result reflects their knowledge, not ChatGPT Search. |
| Brand authority | Wikidata entity and linked profiles, Wikipedia, and asking two models directly whether they know the brand. | Other channels only when reliably known to the model; recommendations. | YouTube, Reddit and LinkedIn are not measured. |
| GDPR review | 9 checks: HTTPS, privacy policy, consent platform, Consent Mode, trackers, US transfers, Google Fonts, controller details. | Fixes for the detected consent platform and trackers. | Documentation (DPAs, records of processing) cannot be verified from a website — it is on the checklist. |
| EU regulations | The GDPR signals plus accessibility statement, AI chatbot, alt text. | Which regulations apply to the business type, findings and a plan. | Company size and sector (NIS2) cannot be determined from a website — listed as "check yourself". |
| Accessibility review | axe-core in a real browser: WCAG 2.1 AA violations by severity and a link to the accessibility statement. | — | Automated checks find roughly a third of issues; keyboard and screen-reader use need a manual review. |
| PageSpeed | Google PageSpeed Insights: Chrome UX Report field data (LCP, INP, CLS) and lab measurements. | — | Low-traffic pages have no field data; the lab score varies by a few points. |
Data and privacy
The tools only read publicly available pages (the home page, robots.txt, llms.txt and the site’s own scripts). An analysis is kept for up to 24 hours so you can unlock the report with your email without running it again. Your email is used to send the report; you only join the newsletter with separate consent. For aggregate statistics (e.g. the share of sites blocking AI search bots) we keep only the domain and technical facts such as bot access and whether a cookie banner is present — no IP address or email. If you opt in to the monthly re-check, we keep your email, the URL and the last score until you unsubscribe (the link is in every email).
Methodology last updated: . Found a problem in a tool? Write to info@createai.si.