Retrieval readiness is the highest defensible state a page can claim after four ordered checks: Meaning, Evidence, Access and Observation. A page can be clear and credible without being accessible to a particular system; it can be accessible without being retrieved; and it can be retrieved without being cited. The framework strengthens the conditions for interpretation and reuse, but it never promises inclusion.
Decision enabled
Decide what the page can claim now, identify the first unmet layer, and choose one controlled next action.
Decision boundary
Do not turn readiness into a universal score, infer authority from schema, or generalise one platform observation to every system, prompt, locale or date.
Status: Founder reviewed by Pero Baldjiev. Founder reviewed by Pero Baldjiev.
The four-layer readiness chain
Evaluate the layers in order. The first unmet layer is the binding failure; later evidence cannot erase it.
Meaning
Are the entities, claims, relationships, conditions and exclusions explicit and internally consistent?
Failure: the page cannot yet be interpreted reliably.Evidence
Can a reviewer inspect the author, sources, method, limitations and correction path behind the claim?
Failure: the page cannot yet be treated as source-ready.Access
Does the page meet the current access, rendering, canonical and visible-content requirements of the assessed surface?
Failure: source quality is not yet platform access.Observation
Has retrieval or citation been recorded with the exact prompt family, surface, locale and date?
Failure: readiness must not be described as observed inclusion.
The five possible states
Report the highest state that the evidence supports. The binding reason explains why the page cannot advance further.
Meaning blocked
Meaning or evidence is not sufficient for source readiness. Record whether the binding reason is ambiguity, contradiction, missing support or missing accountability.
Source-ready
Meaning and evidence pass. Platform-specific access is unverified, failed or outside the current assessment.
Access-ready
Meaning, evidence and access pass. No bounded retrieval or citation observation has yet been recorded.
Retrieval observed
The page or a passage was surfaced under a recorded test. This is not evidence that it will appear for every user or prompt.
Citation observed
The page was shown as a cited or linked source in the recorded surface. This does not prove universal authority, preference, traffic or conversion.
Confidence: Low, Moderate or High must be reported separately. High confidence requires current evidence, repeatability, adequate breadth and no unresolved contradiction.
Completed worked example: a procurement answer page
Illustrative fictional composite. This example demonstrates the decision rule without representing a published client result.
A technically indexable page answers a complex procurement question, but it uses inconsistent product names, hides the author, buries the direct answer and provides no source trail.
| Layer | Evidence | Decision |
|---|---|---|
| Meaning | Product, organisation, method and conditions now resolve consistently; the direct answer preserves its limitations. | Pass |
| Evidence | An accountable author, primary sources, method note and correction path are visible. | Pass |
| Access | The page is canonical and indexable, but the target answer surfaces have not yet been checked under a recorded protocol. | Not verified |
| Observation | No prompt, surface, locale or date record exists. | Not observed |
State: Source-ready
Confidence: Moderate
Binding failure: Access has not been verified for the assessed surfaces.
Permitted action: Run platform-specific access checks, then begin a bounded observation sample.
Remaining uncertainty: Retrieval and citation have not been observed.
Verify the first unmet layer. Do not add more schema, rewrite for imagined model preferences or claim visibility before the access and observation records exist.
Run the full assessment
The fast path above is sufficient to classify the current state. Open only the detail needed for the next layer.
How to apply the four layers
- Define the assessed page and decision. State the page, claim, audience, market and platform surface.
- Resolve meaning. Stabilise entity names, direct answers, relationships, conditions and exclusions.
- Qualify evidence. Expose authorship, sources, method, review status, limitations and correction paths.
- Verify access. Check the requirements that apply to the named surface rather than assuming one crawler or index represents every system.
- Observe. Record exact prompts, surfaces, locales, dates and cited URLs.
- Improve cautiously. Change the page for reader value and repeated evidence, then preserve the previous state and rerun trigger.
Platform-specific access checks
Keep separate records for each assessed platform. Confirm only what the current official guidance and direct tests support.
- Search and AI search surfaces: record crawlability, indexability, canonical handling, rendering and any surface-specific controls.
- OpenAI search access: distinguish search access controls from other crawler or training controls.
- Structured data: verify that markup matches visible content and page purpose. Treat it as context, never proof of authority or inclusion.
- Unavailable evidence: mark the access layer unverified rather than inferring a pass.
Observation protocol and evidence record
| Field | Record | Why it matters |
|---|---|---|
| Page or passage | Exact URL and relevant passage | Prevents evidence being transferred to another page. |
| Prompt sample | Exact prompt or bounded query family | One prompt cannot represent all demand. |
| Surface | Named product and result type | Retrieval and citation can differ by surface. |
| Context | Locale, language, date and signed-in state where relevant | Results can vary by context and time. |
| Outcome | Not observed, retrieved, cited or linked | Keeps readiness separate from inclusion. |
| Evidence | Screenshot, export or reporting reference | Allows another reviewer to challenge the observation. |
Absence from a limited sample is not proof of universal exclusion. A single citation is not proof of universal preference.
Advanced evidence requirements
- Store the observation date with the platform guidance used at that time.
- Separate visible page facts, technical test results, inferred readiness, observed retrieval, observed citation and commercial outcomes.
- Record contradictory observations rather than averaging them away.
- Reduce confidence when evidence is stale, narrow, non-reproducible or disputed.
- Preserve the prior state, reviewer and reason whenever the conclusion changes.
Messy reality: platforms disagree
Illustrative fictional composite. This scenario tests conflicting platform observations and reviewer disagreement without representing a published client result.
A research page is technically accessible and well sourced. Google shows it as a supporting link for one query family, Bing records several citations in one market, and a bounded ChatGPT search sample does not cite it. One important claim is still ambiguous, and two reviewers disagree about whether the page is “ready.”
State: Meaning blocked
Confidence: Low while the material ambiguity and reviewer disagreement remain.
Binding failure: One important in-scope claim is ambiguous, so the page cannot advance beyond the Meaning layer.
Bounded observations: Retrieval or citation was observed on the named Google and Bing surfaces, but those observations do not override the Meaning failure.
Permitted claim: Limited retrieval or citation was observed on the named surfaces under the recorded tests.
Prohibited claim: The page is universally citation-ready, preferred by answer systems or guaranteed to produce traffic or conversion.
Action: Resolve the ambiguous claim, preserve the platform-specific observations, rerun the assessment and record reviewer disagreement.
Decision discipline: Systems can disagree. Readiness is not inclusion; inclusion is not authority; citation is not a commercial outcome.
Retrieval Readiness Check
Use the five fast questions to identify the current state. Advanced controls refine confidence; they do not override the binding-failure rule.
Claims, failure rules and reopening
Permitted and prohibited claims
Permitted
- State the highest layer supported by the evidence.
- Name the exact surface, prompt family, locale and observation date.
- Describe a page as source-ready or access-ready when the relevant checks pass.
- Report bounded retrieval or citation observations with separate confidence.
Prohibited
- Promise inclusion, citation, ranking or conversion.
- Call schema, chunking or concise formatting a universal platform requirement.
- Transfer one observation to every system, user, market or date.
- Describe a proprietary state as externally validated without evidence.
Failure and refusal rules
- Schema substitution: markup is added while visible meaning or evidence remains weak.
- Answer fragmentation: a page is split into shallow blocks that remove conditions or repeat commodity information.
- Entity inconsistency: names, roles or framework labels change across the evidence set.
- Observation inflation: one appearance is presented as durable or universal visibility.
- Refusal condition: when required evidence is missing, contradictory or outside scope, narrow the claim or stop.
Review and reopening triggers
- Official platform guidance materially changes the access conditions.
- Structured data no longer matches visible content or page purpose.
- The source, author, method or evidence becomes unreliable.
- Observed patterns cannot be reproduced across the stated sample.
- Content changes improve extraction at the cost of reader clarity or necessary context.
- A new market, locale, surface or decision boundary is introduced.
Retain the previous state, changed evidence, reviewer, date and affected decision.
Reviewer governance
- Review the first unmet layer rather than debating a composite score.
- Record material disagreement and reduce confidence until it is resolved.
- Require another reviewer to reproduce the observation record before assigning High confidence.
- Keep founder-review status separate from authorship and publication status.
Connect readiness to the wider decision system
Use adjacent frameworks only for the next unresolved decision; do not repeat their methods inside this article.
Authoritative context and review sources
- Google Search Central — AI features and your website
- Google Search Central — Optimizing for generative AI features
- Bing Webmaster — AI Performance public preview
- OpenAI — Overview of OpenAI crawlers
These sources define external platform context and eligibility boundaries. They do not certify the proprietary SERPIFAI framework or validate an unstated outcome.