How Everfound works

We fetch your homepage and up to 30 pages (your selection, or our auto-pick of the pages that matter: pricing, FAQ, about, comparisons), plus the machine files AI crawlers use: robots.txt, sitemap.xml, and llms.txt. We parse any JSON-LD structured data, and we report coverage honestly: “audited X of N pages found.”

We take the questions real people type when they need what you offer and ask them across six AI engines: OpenAI GPT-5.5, Anthropic Claude Opus 4.8, xAI Grok 4.3, Google Gemini 3.5 Flash, and Perplexity Sonar Pro with live web search, plus Google AI Overviews read from live Google results. We ask each engine every question once per run, and read the aggregate across questions, engines, and runs rather than any single answer. For every answer we record whether you were mentioned, where you ranked, how you were framed, and which sources the engine cited. The report keeps the receipts: the actual answers and the exact links, so nothing here is our word against the machine's.

Findings and the action plan are grounded in what the crawl actually saw. The kit is evidence-aware: anything your site already does well (existing schema, an existing llms.txt, an existing FAQ) is acknowledged and extended, never replaced. Intentionally built content stays yours. We suggest merges, not rewrites.

The score out of 100 weighs six categories:

• Entity clarity (20%) · can a machine tell what you are, who you serve, and why it should recommend you.
• Crawlability (18%) · can AI crawlers reach and read your pages, including robots.txt, sitemap.xml, and llms.txt.
• Structured data (18%) · the JSON-LD that fits your kind of site.
• Owned content (16%) · the pages assistants quote: FAQ, pricing, about, comparisons.
• AI answer presence (16%) · the live measurement above.
• External consensus (12%) · the third-party footprint AI trusts.

Three properties keep the score honest. Every check shows its evidence: the report tells you why each point was earned or missed, never just a number. The on-page half is deterministic: the same site with the same facts scores the same, every time. And checks only apply where they're fair: a law firm isn't graded on app-store presence, a web-only product isn't penalized for skipping Google Play, and a check we didn't run (like llms.txt on an older crawl) is omitted rather than counted against you.

The on-page half of your score is stable. The live half measures a moving world: AI engines sample their answers, their web indexes shift daily, and the same question can surface different sources on different days. A mention count that reads 5/17 one day and 4/16 the next is normal variance, not a regression. It usually means one engine answered differently, or one call returned nothing (we drop failed calls rather than invent them). That's why we show the receipts and judge the trend across runs rather than any single point.

A re-audit re-crawls your site fresh, re-asks the same questions, and appends a new run to your history. It never overwrites anything. On-page fixes register immediately: publish schema, add missing pages, fix readability, and the deterministic categories move the same day. The report then shows the comparison, check by check: the ones that now pass, the ones that improved, and the ones that lost ground. Every previous run stays retrievable, so the before and after is provable, not anecdotal.

• Same re-audit: structured data, llms.txt, new pages, content fixes, entity clarity. The crawl sees them and the score moves immediately.
• Days to weeks: AI mentions, rankings, and citations move on the engines' timetable, not yours. They re-crawl and re-index on their own schedule. The repairs make you recommendable; the engines decide when to notice. This lag is real and we won't pretend otherwise.
• Compounding: the “Where AI looks” list in your report names the third-party sources the engines actually cited for your category. Earning presence there is the long game that moves the live half.

Because the live half moves on the engines' schedule, monitoring keeps measuring it for you. Weekly monitoring re-asks up to 3 of the questions you track across all 6 AI assistants, watches your pages for changes, and emails you when something moves: a lost mention, your share of mentions slipping, or an assistant still repeating something you took down. It does not re-crawl or re-score your site. That is what a re-audit does, so the on-page half of your score moves when you run one, not on the weekly schedule. Monitoring is priced per site and runs on a plan. A one-time audit measures a single point in time and includes a second run to check your fixes worked. Cancel anytime. The weekly runs build the trend lines that turn one-off measurements into proof.

We designed the audit so you never have to take our word for anything. The visibility section links the actual AI answers and the exact pages they cited. Every score check shows its evidence. Every run is kept, so improvement is a measured delta, not a promise. The honest claim is this: we repair the inputs AI systems act on (readability, clarity, structure, citability), then measure weekly whether the engines respond. When the mentions come, you'll see exactly where, in which engine, framed how, and against which competitors. If something we recommended didn't move the needle, the same receipts will show that too.