What shipped, and when

Every free audit now asks Claude your buyer's question with live web search, then shows the answer itself.

• The verdict, observed: cited, found but not cited, or not found, read from the answer and its sources.
• Every source, clickable, plus the pages Claude's search read but didn't cite, and which sites the citations came from.
• Brand mentions: whether the answer names you at all, and how early.
• Relevance as a distance: your passages and the cited pages are matched with Google's Gemini embeddings. The comparison stays labelled as a simulation.

One answer is one sample; the result page says so, with the model and the time.

Console probes ask every answer engine that has a key, twice, and measure how often each cites your page, with its uncertainty.

• Engines: Claude on every probe; ChatGPT search, Perplexity, Google AI Overviews and Copilot where their keys are set, or with your own keys. Gemini is not measured: Google's terms don't allow it.
• The plan and the Fix Kit are written by Claude from the measured gap and your page. When a stage can't run, the report says why instead of filling the gap.
• Recent probes: reopen any report, and re-check after you ship a fix.
• Monitor: each watched question keeps a dated history and the latest answers, and changes show up when an engine starts or stops citing you.

The whole site and the console are rebuilt around how an answer engine actually works: it looks pages up, reads a few, cites fewer.

• The audit form sits on the homepage, and real checks replay beside it.
• Result pages lead with the observed verdict, then the answer and its sources, then the readiness signals and the fixes, each with a Copy button.
• One navigation everywhere, built to work as well on a phone as on a desktop.

• Free work requests are limited to 2 a minute per visitor, so the free audit stays free for everyone.
• The optional follow-up email on a free audit is never part of the public result page.
• Tenant data is isolated at the database level, with every query scoped to its account.

Signal expansion grounded in published AEO research.

• BLUF (Bottom-Line-Up-Front) detection.
• Structural depth scoring (GEO-SFE methodology from arXiv:2603.29979).
• Wikidata brand-entity validation at probe time.
• llms.txt validator + generator per the llmstxt.org spec.
• Sub-query coverage scoring.
• Mega-page coverage signal.
• YouTube embed weighting per Ahrefs 2025 research.

Site-wide multi-page audit, query expansion, and closed-loop schema verification all shipped together.

Added E-E-A-T as a dedicated rubric category covering first-person evidence, case study evidence, author credentials, byline depth, brand entity, press mentions, privacy/terms, and contact info.

Wikidata brand-entity check runs at probe time: sites without a verified brand entity fail the trust anchor signal that engines use to validate identity.

Audit result pages now cache at the CDN edge for 24 hours per slug. Viral shares hit the CDN, not the database.

Cold-start UX: visible 5-stage progress meter during longer probes so users see real work happening (crawling → probing → embedding → computing → building fix kit).