How we measure

The AI-Readiness score (0–100)

We fetch your page's raw HTML — the same no-JavaScript view an AI crawler gets — and score eight weighted dimensions, including a quotable answer near the top, entity clarity, valid structured data, heading structure, machine-readable freshness, and whether the content exists in raw HTML at all. Each dimension shows the exact evidence behind its points, so you always see why a page scored the way it did.

The score is recomputed from the live page on every check and goes down if the page regresses. A page we cannot reach gets no score — an outage is unknown, not zero. Every check is kept as history, so before/after deltas around your fixes are real evidence, not memory.

The crawler-access check

We read your robots.txt and evaluate it against the current, documented AI crawler user-agents from OpenAI, Anthropic, Perplexity, Google, and Common Crawl — a list we keep up to date as vendors change their crawlers. We distinguish live crawlers (fetch pages to answer users now — the citation path) from training crawlers (affect future models). We also report whether an llms.txt exists — as information only, because engine adoption of that convention is unproven. We check for blanket root blocks; path-level rules are out of scope and we say so rather than overclaim.

Change verification

When you mark a fix, we snapshot the page's readiness dimensions. "Verify change" re-reads the live page and reports which dimensions actually improved — binary, on-demand, no AI model involved. If nothing changed, we tell you that too: it usually means the change did not deploy.

AI citations — measured as a rate, never a yes/no

AI assistants are non-deterministic: the same question can cite your site in one run and not the next. A single scan is noise. We therefore report citations as "cited in N of the last M scans" with a stated confidence tier (fewer than 3 scans = low, 3–5 = moderate, 6+ = good). A new citation is a leading indicator worth knowing about the moment it happens — it is not the KPI.

Proof of impact — honest windows

Search results move slowly. We read an early signal from about 2 weeks after a fix and label it provisional; the verdict comes at 8–12 weeks. Nothing auto-confirms before the verdict window — "no movement yet" inside it means on schedule, not failure. To separate your fix from site-wide drift (algorithm updates, seasonality), we compare the fixed page against the median movement of your unfixed pages over the same window. Regressions are shown with the same prominence as wins, and attribution is always stated as correlation, not causation.

What we do not promise

Whether ChatGPT, Claude, Perplexity or Gemini ultimately cites your site is decided by those models and by your site's off-site authority. Nobody controls that — not you, not us, not any tool. What SEO Sets promises is the part that can be promised: your pages measurably ready to be cited, your crawler access open, your changes verified live, and every outcome — including "this didn't work" — reported truthfully with the next step attached.