My AI Readiness Score Dropped — Why?
A lower number is an observation to investigate, not a diagnosis. Before assigning a cause, confirm that the two results describe the same target under a sufficiently comparable score version, coverage state, and evidence basis. If those conditions differ, subtracting one score from the other may create a false story.
Start with the exact model
ClickRadius AI Readiness estimates five weighted evidence families: Off-site Authority (28%), On-page GEO (26%), Schema Engine (18%), Entity / E-E-A-T (18%), and Crawler Access (10%). These are current starting estimates, not weights fitted to customer outcomes. SEO Health and monthly sampled AI Presence remain separate.
The composite is withheld when supported evidence does not clear the versioned minimum coverage threshold. Missing evidence remains unknown rather than becoming zero. A value of zero, an unavailable family, and a family measured from stale evidence are therefore different states.
Five checks before explaining a change
1. Did the target identity change?
Compare the canonical organization, submitted URL, host redirects, and scan scope. A homepage result and a different subdomain or redirected destination are not the same measurement merely because the brand name looks familiar.
2. Did the score version change?
Only compare results produced under a compatible model. A historical score from a retired family definition cannot be subtracted from the current five-family score as if the scale were unchanged. Preserve the model identifier with every retained result.
3. Did evidence coverage or applicability change?
Compare pages seen, measured coverage, family applicability, evidence age, and insufficient-data reason. A newly unavailable source can change what the model is able to support even when nobody deliberately edited the site. It must not be described as a measured deterioration without inspecting the underlying evidence.
4. Which exact family evidence changed?
Inspect the before-and-after family envelopes and their findings. An Off-site Authority observation is not interchangeable with Entity / E-E-A-T; On-page GEO is not Schema Engine; and Crawler Access reports declared policy rather than an observed external crawl. The composite alone cannot identify which evidence moved.
5. Could the measurement or processing path be wrong?
Do not rule out a defect simply because a score exists. Review fetch and parse evidence, timestamps, error states, source receipts, and repeatability on the same controlled input. If those records disagree, report the disagreement and investigate it instead of constructing a content or market explanation.
A score delta is defensible only after identity, version, coverage, and family evidence are comparable. Until then, the cause is unknown.
— ClickRadius Institute
If the comparison is valid, inspect supported changes
- Off-site Authority: review which directory, review, syndication, social, link, and entity-record observations changed or became unavailable.
- On-page GEO: compare the actual passages, structure, specificity, sources, and promotional balance observed on the measured pages.
- Schema Engine: compare parsed structured data with the visible facts it describes; syntactic validity and factual support are separate checks.
- Entity / E-E-A-T: inspect entity completeness, author identity and supported credentials, and agreement between owned and independent sources.
- Crawler Access: compare the site's declared robots policy and
llms.txtobservation. Do not turn policy into proof that an engine crawled the site.
From diagnosis to governed correction
A supported finding can justify preparing an exact revision. Applying that revision requires current authorization and a connection proven capable of that effect; otherwise it remains for review or customer implementation. Retain the provider response and perform an independent public-origin reread where supported. Provider acceptance is not itself a public effect.
A later AI Presence change remains separate. ClickRadius samples ChatGPT, Gemini, Perplexity, Claude, and Grok monthly; Microsoft Copilot is not currently supported. The sample can show what those configured observations returned, but neither its timing nor a readiness delta proves what caused an engine answer.
Frequently asked questions
My score dropped but I did not change the site. How is that possible?
First confirm that both results used the same target, score version, and sufficiently comparable evidence. Coverage, evidence availability, off-site observations, or the measured public surface may differ even when no deliberate edit was made. A processing defect must also remain a possibility until the retained evidence rules it out.
Which family should I fix first after a drop?
There is no universal answer. Review unknown and stale evidence, the size and support for each family delta, dependencies, risk, authority, and connector capability. Crawler Access describes declared policy; a lower value does not prove that an external engine lost access.
Does a lower AI Readiness Score mean citations were lost?
No such conclusion follows from the score. AI Readiness estimates setup evidence. AI Presence separately reports monthly sampled answers, and neither a score change nor a later observation establishes causation. External engines control their own crawl, index, ranking, recommendation, and citation decisions.
Use a free AI Readiness assessment to inspect supported evidence at a submitted URL, or see ClickRadius plans for governed correction preparation and separate monthly AI Presence observations. Compare only compatible score versions and evidence states.