Research · Limitations
AI SEO audit limits: fetched pages, engine inputs, scores, and citations
In brief. This is the central limitation of the aiseo-audit project on GitHub: Agency Enterprise's tool deterministically reviews fetched HTML, page inputs, and public site signals. It does not inspect a private AI engine index, retrieval model, hidden context, or citation policy, so the score measures readiness only.
Maintained by Jeff Patterson and Agency Enterprise · Updated August 30, 2026
Measurement boundaries: overview
The boundary matters because a page audit and a deployed AI search engine observe different systems. This means the tool reports only the inputs it fetches and the rules it applies.
Fetched HTML, not a browser-rendered page
The audit does not run client-side JavaScript. Content added after page load stays outside the result.
No visual or multimodal assessment
It cannot judge layout quality, intrusive advertising, screenshots, or information carried only in pixels.
Public HTTP access is assumed
Authentication, consent flows, bot mitigation, personalization, and geography change what the server returns.
Fetched inputs vary
Identical inputs produce the same score. Redirects, experiments, dynamic HTML, and network failures change those inputs.
No proprietary engine visibility
The audit cannot inspect a commercial engine’s index, retrieval model, hidden context, or citation policy.
No citation guarantee
Readiness signals do not predict retrieval or citation by a deployed engine for a particular answer.
AI SEO audit limitation terms
Fetched HTML
Fetched HTML refers to the response body returned to the audit request.
Readiness score
A readiness score refers to the weighted page signals observed by one tool version and configuration.
Citation forecast
A citation forecast refers to a model-generated percentage for a future engine citation.
Private index
A private index is defined as the engine-owned collection of content unavailable to the audit.
Deterministic result
A deterministic result refers to the same score produced from the same fetched inputs and settings.
Observed citation
An observed citation is a type of separate engine measurement recorded for a specific query and time.
How to interpret score values
| Value | Meaning | Limit |
|---|---|---|
| 100% stage score | Every scored factor in that stage earned full points. | It is not a 100% citation rate. |
| 90% page score | The page meets the current A-grade threshold. | It is not a 90% traffic or citation forecast. |
| 0/0 observation | The observation stays visible outside the score. | It neither adds nor removes points. |
How to interpret a score
Compare the same page or stable page set under the same tool version and configuration. A score change means the observable inputs changed. It does not prove that traffic, rankings, or citation odds changed. The reason is that the audit does not observe those outcomes.
Audit points are internal weights. Apply a recommendation, review the page as a reader, and measure again. For continuous integration, set a threshold from your own baseline. No universal score is a scientific pass mark. This works by treating the project's earlier result as the comparison point. As a consequence, the gate detects regressions without pretending to predict citations.
Responsible use
- Keep factual quality and reader usefulness ahead of score maximization.
- Do not add stale dates, superficial citations, or repeated terms to chase points.
- Review conditional findings in the domain and pipeline stage where they were tested.
- Use target queries that reflect genuine audience needs.
- Preserve a baseline whenever upgrading a major tool version.
Start with one page and the queries it is meant to answer.
Key takeaways
- The audit is a measurement of fetched page and site signals.
- The score is not a prediction of traffic, ranking, or citation rate.
- The engine is a separate system with private retrieval and citation rules.
- The comparison is valid only when the page inputs, settings, and major version stay stable.
- The baseline is the right reference for a project-specific quality gate.
Bottom line: use the score to compare observable page readiness, then record real search and citation outcomes separately.
Sources
According to OpenAI's API documentation, model outputs are variable by nature [1]. According to a 2026 ACL study, generative search systems differ in the sources they use[2]. According to the project evidence map, the audit separates supported factors from conditionals, heuristics, diagnostics, and experiments[3].
According to the GitHub documentation, aiseo-audit analyzes fetched page and domain signals [4]. According to the version 2 migration guide, major scoring changes require new baselines [5].
- API backward compatibility, OpenAI.
- Characterizing Web Search in the Age of Generative AI, Findings of ACL 2026.
- aiseo-audit evidence map, project research record.
- aiseo-audit documentation, GitHub.
- Version 2 migration guide, project documentation.