AI Discoverability Audit
How AI systems
see your content
Five pillars, scored 0 to 4 against evidence. You get a scorecard, the artefacts behind every score, and a remediation backlog.
Delivered in partnership with NarrativAI
Five pillars
Accessible
Feeds, crawler and agent access, retrievability, declarations.
Accurate
A 120-prompt sweep across fixed surfaces, three runs each. Extractability, source-of-truth alignment, sentiment.
Consistent
Divergence across every retrievable source carrying your data. Propagation routes, third-party content.
Chosen
Agent selection testing against a named competitor set. Attribute flips, price sensitivity, verifiable authority.
Observed
Server-side bot logs, crawl and citation coverage, the agent funnel, alerting.
Red team
Time to block, volume before block, cost to extract your catalogue. Crawler spoofing and headless detection. On the defensive side, which legitimate agents and partners your current controls block. Under signed rules of engagement.
Scoring
Check scores
| Score | Meaning |
|---|---|
| 0 | Not in place |
| 1 | Partly in place, not managed |
| 2 | In place but incomplete or inconsistent |
| 3 | Correct, complete, documented |
| 4 | Correct, complete, and monitored with alerts |
Overall bands
| Score | Band |
|---|---|
| 0–24 | Exposed |
| 25–49 | Reactive |
| 50–69 | Managed |
| 70–84 | Instrumented |
| 85–100 | Directed |
Pillar scores are reported alongside the overall number.
Critical failures
These cap the overall score at 40 and go on the front page of the scorecard:
- — A major crawler blocked or erroring on important pages by accident
- — Wrong price, stock, or regulated attribute in feeds or on product pages
- — An AI assistant repeating an error with legal or safety exposure
- — No server-side bot logs available
- — Crawler configuration contradicting your licensing position
Three outputs
Scorecard
Overall score, band, five pillar scores, critical failures, trend against your last audit. Control-area breakdown on page two.
Evidence
Access matrices, raw prompt responses and citations, the divergence matrix, selection test results, bot heatmaps, red team logs. Dated and reproducible.
Remedies
Each finding as a ticket: severity, impact, remedy, owner, effort, and how it gets verified.
Platform leverage brief
A separate document for organisations with significant AI platform spend: grounding and citation manifests, which sources influenced an answer at inference, adjacency controls, targeting transparency.
Delivery
About 26 working days, kickoff to alignment session.
| Days | Phase | Work |
|---|---|---|
| 1–3 | Scoping | Markets, templates, competitor set, prompt bands, access requests |
| 4–12 | Collection | Testing, crawls, log ingestion, feeds, source register |
| 8–15 | Sweep | Prompt sweep and sentiment analysis |
| 10–18 | Selection testing | Agent harness runs, attribute flips, credential landscape |
| 12–17 | Red team | Adversarial testing, offensive and defensive |
| 10–17 | Tooling | Tool review, interviews, provenance |
| 16–22 | Analysis | Scoring, divergence matrix, root cause, benchmark |
| 22–26 | Delivery | Scorecard, evidence, backlog, alignment session |
Re-run quarterly or twice a year. Prompt set, surfaces, competitor set and weights stay fixed between runs. Access and accuracy checks can run continuously in between.
Request an audit conversation
Tell us a little about your organisation and what you want to understand. Your enquiry will be shared with NarrativAI so the team can determine whether an AI Discoverability Audit is appropriate.
Your details are used to respond to this enquiry and shared with NarrativAI for that purpose. Privacy.