Generative engine optimisation
Find out what GPT · Claude · Gemini · Perplexity can actually read on your site.
Paste a URL. We fetch it live, parse what a retrieval system would parse, and return a computed 0–100 score across ten pillars with a ranked list of what to fix first. No estimates, no simulated data, one free run.
One free audit. No signup. Roughly 40 seconds — we make the requests live.
—
audits run
—
median score so far
—
pillars failed per site
10
pillars scored live
What we check, and why each one costs you citations
12%
Crawlability & AI bot access
Seven named agents, each with a different job. Blocking OAI-SearchBot removes you from ChatGPT's answers, not from its training set.
8%
llms.txt & agent readiness
A curated map for agents that already have access. Voluntary, unproven, and half an hour of work — weighted honestly.
13%
Structured data
JSON-LD is the only unambiguous way to state who publishes this, who wrote it and when it changed.
15%
Content extractability
Retrieval lifts passages. If your answer needs the three paragraphs above it, it does not survive extraction.
13%
Evidence & E-E-A-T
Named authors, dated figures, linked sources. Unsupported superlatives lose to a competitor who published a number.
8%
Entity authority
Wikidata and Wikipedia matching. Without a node in a public graph you are a string, not a thing.
10%
Off-site citation sources
Reddit mentions, live. LinkedIn and Quora have no free API, so we hand you the exact search and label it manual.
10%
Local & reputation signals
Google Business Profile completeness, review volume and recency — skipped and redistributed when no key is connected.
6%
Multimedia & accessibility
Alt-text coverage and VideoObject schema. What a text pipeline reads when the meaning lives in a picture.
15%
AI share of voice
Category prompts run live against Gemini, mentions classified for sentiment. Reported, not scored, until more engines connect.
Weights sum to 110 before normalisation because share of voice is reported, not scored, until more engines are connected. Read the full methodology.
Example report — illustration only
What the report looks like
The panel on the right is a fixed sample using invented numbers, shown so you know what you are getting before you spend 40 seconds. Everything on your own report comes from live requests to your URL.
Illustrative figures. Not a real audit.
Questions we get from people who don't believe us yet
- Is the score real, or a marketing number?
- Every pillar is computed at request time from live data: we fetch your robots.txt, llms.txt, sitemap and HTML, parse your JSON-LD, query Wikidata and Wikipedia, search Reddit, and put your rendered text in front of Gemini. Nothing is cached or estimated. Where a check cannot run, we mark it unavailable and redistribute its weight instead of guessing.
- Which AI engines do you actually test for share of voice?
- Gemini only, on the free tier. ChatGPT, Claude and Perplexity have no free API, so we do not test them and we do not model them. Because only one engine is testable, share of voice is reported but excluded from the score, and its 15% is redistributed across the pillars we can measure. Connect your own keys on the paid tier to include the other engines.
- Why did my score change when nothing on my site changed?
- Two pillars are judged by a language model and one depends on third-party mentions. Model answers shift as engines re-crawl and retrain on their own schedule. Treat a single audit as one reading; the trend across monthly runs is the signal.
- Will fixing these ten pillars get me cited by ChatGPT?
- It removes the reasons you are being skipped. Access, structure and evidence are prerequisites — nobody gets cited without them. Whether a specific answer names you also depends on off-site consensus you do not control, which is why we weight Reddit and entity signals at 18% combined and tell you when the work is a quarters-long job rather than an afternoon.
- Do you store my site's content?
- We store the report: scores, findings and the action plan. We do not keep a copy of your page HTML after the audit finishes.