Steal this: the sixty-second AI citation check I run before any strategy call — then expand it into a proper white-label audit your team can repeat monthly.

If you pitch Answer Engine Optimization without knowing whether the brand is cited, ignored, or misrepresented, you are selling blind. This checklist keeps the work qualitative, honest, and free of fabricated share-of-model statistics.

Why audit before strategy

Discovery calls fill with assumptions:

• “We’re invisible in AI.”
• “Competitors own ChatGPT.”
• “We need 50 new blogs.”

Sometimes those statements are directionally true. Sometimes the brand is cited but for the wrong queries. Sometimes the model recommends a founder who left three years ago. A citation audit replaces folklore with a log.

The 60-second pre-call check

Use this when you have almost no time:

1. Open your notes doc.
2. Write five prompts a buyer would actually ask (not your favorite head terms only).
3. Run each in ChatGPT and Perplexity (same day, same wording).
4. For two commercial prompts, check Google for an AI Overview if it appears in your market.
5. Mark: cited / not cited / partial; note who else appears.
6. Walk into the call with one insight sentence: “You’re absent on hire-intent prompts; present on definitional prompts; competitor X owns comparisons.”

Sixty seconds per prompt surface is unrealistic if you dawdle — batch fast, refine later. The point is a directional pulse, not academic perfection.

Full white-label audit framework

Step 1 — Freeze the prompt set

Build 15–25 prompts across clusters:

• Problem aware (“how to reduce X”)
• Category (“best Y for Z”)
• Comparison (“A vs B”)
• Vendor (“who should we hire for…”)
• Risk (“what goes wrong when…”)
• Local modifiers if relevant

Version it: Prompt Map v1.2 — 2026-09-22 . Do not silently edit mid-month.

Step 2 — Surfaces and cadence

Minimum surfaces:

• ChatGPT (note settings/browse mode if it changes behavior)
• Perplexity
• Google AI Overviews when shown for the query/locale

Optional: other engines your ICP uses. Monthly cadence for retainers; weekly during rebuild sprints.

Step 3 — Logging columns

| date | prompt | surface | cited | brand_or_url | competitors_cited | evidence_link_or_note | gap_type | next_action |

Gap types: no_entity, weak_answer_page, stronger_competitor_proof, outdated_entity, off_niche_citation, hallucination_risk.

Step 4 — Competitor compare

Do not only ask “are we cited?” Ask “who is the default recommendation?” Capture two competitors every cycle. Patterns matter more than single lucky wins.

Step 5 — Site correlation

For each miss, map to:

• Is there an owned URL that should answer this?
• Is the answer buried?
• Is the entity unclear?
• Is there zero third-party corroboration?

This correlation turns the audit into a backlog, not a anxiety report.

Step 6 — Executive summary format

One page:

• Coverage snapshot (counts, not fake %)
• Three wins
• Three losses
• Top five backlog ships
• Risks (misrepresentation, outdated bios)

What never to do in an audit

• Invent a “37% share of ChatGPT” claim from manual checks
• Change prompts until you get a flattering screenshot
• Promise that fixing one page guarantees citations
• Ignore hallucinations — note them carefully without amplifying misinformation
• Publicly attack competitors with scrapes; keep evidence internal

Agency packaging

Productize as:

• Diagnostic (1–2 weeks): map + baseline log + backlog
• Monthly monitor: log refresh + slide
• Rebuild sprint: answer-shaped pages tied to worst gaps
• Mention sprint: podcasts/roundups for vendor prompts

Price the diagnostic so it can stand alone. Many clients need truth before a six-month retainer.

Delivery context: services.

Integrating Search Console generative data

When generative AI reports appear in Search Console for a property, add them as a companion instrument. Use Google’s definitions. Do not force-fit them into your manual citation percentage cosplay. Qualitative logs still explain which questions you win or lose.

Proof pattern (anonymized)

A partner ran the sixty-second check live on a sales call. The prospect’s brand appeared for definitions and vanished for “who should we hire.” Competitors with podcasts and founder pages dominated. The deal closed on a diagnostic, not a vague AEO promise. Month one rebuilt hire-intent pages and updated entity bios; the log became the monthly ritual. No fabricated rates — just a visible backlog burning down.

Team training tips

• Record a Loom of one full audit once; reuse for onboarding.
• Pair juniors with seniors for hallucination judgment.
• Ban vanity screenshots in Slack without log rows attached.
• Review gap_type taxonomy quarterly so tags stay clean.

Ethical and legal notes

Respect platform terms. Do not scrape behind authentication walls or automate abuse. For regulated clients, route public-facing responses through compliance when audits reveal dangerous misinformation that you might publicly correct.

Sample prompts (illustrative — rewrite for your category)

• “What should a mid-market company evaluate before hiring an SEO agency in 2026?”
• “SEO vs AEO vs GEO — how do they work together?”
• “How do AI Overviews change reporting for agency clients?”
• “Who are credible experts on answer engine optimization for agencies?”
• “What is an answer-shaped page?”

Replace with client language. The quality of the audit equals the quality of the prompts.

Scoring rubric (qualitative)

For each prompt/surface:

• 0 — Absent or wrong entity recommended
• 1 — Tangential mention without useful citation
• 2 — Cited among others with partial accuracy
• 3 — Clear, accurate citation or recommendation

Average carefully — and never publish averages as industry benchmarks. Use scores to prioritize backlog only.

After the audit: workshop agenda (90 minutes)

1. Review coverage snapshot (15)
2. Live-run three painful prompts (20)
3. Map gaps to URLs and entities (25)
4. Choose sprint ships (20)
5. Confirm reporting cadence and owners (10)

Leave with owners and dates. Audits without owners become PDFs that die in Drive.

Tooling notes (keep humble)

Spreadsheets still win for many teams. If you use monitoring software, validate samples manually each month. Tools drift. Interfaces change. Human judgment on hallucination and brand safety cannot be fully outsourced.

Store evidence privately. Client decks get summarized patterns. Oversharing raw model outputs can create legal or reputational oddities if content is sensitive.

Misrepresentation protocol

If a model attributes a false claim to your client:

1. Log it
2. Check owned pages for ambiguous wording that could cause the error
3. Strengthen clear corrective answer pages
4. Improve entity clarity
5. Consider public clarifications carefully with legal
6. Do not amplify the false claim in your own marketing

Audits sometimes surface reputation issues. Treat them seriously.

Scaling audits across twenty clients

Without a system, audits die.

• Shared template workbook with tabs per client
• Prompt maps stored beside SOWs
• Calendar reminders for monthly refresh
• Spot-check QA by a senior on 10% of rows
• Library of anonymized insights for training (never leak client secrets)

Productized operations beat heroic analysts. The checklist is the product; the analyst is the steward.

Connecting audits to revenue conversations

End every audit readout with: “If we close these three gaps, which sales conversations get easier?” Force the bridge. Citation work that never touches revenue language becomes another vanity report — exactly what we are trying to escape.

Junior analyst onboarding script

Day one: read this checklist.
Day two: shadow a senior on five prompts.
Day three: solo log ten prompts; senior reviews gap tags.
Day four: draft an executive summary page.
Day five: present to an internal AE as practice CMO.

By Friday they can contribute without inventing metrics. That is the bar.

Field note

Keep the standard high and the tone calm. Agency buyers have seen enough chaos marketing for one decade. Your advantage is sequenced delivery, honest instrumentation, and language a CMO can repeat in their own leadership meeting without flinching. That is the work. That is what renews.

FAQ

How many prompts are enough?

Fifteen to twenty-five for most mid-market audits. Enterprise categories may need segmented maps per product line.

Will results differ by user and time?

Yes. That is why patterns and repetition matter more than one screenshot. Note major product changes when interfaces shift.

Should we include Bing, Gemini, or others?

Include what your buyers use. Start with the big three surfaces above; expand deliberately.

Can we white-label the slides?

Yes — keep methodology honest. Do not let white-label branding become an excuse for fake metrics.

What is the first ship after a bad audit?

Usually: entity cleanup + one answer-shaped page for the highest-intent miss — not fifty thin blogs.

Reminder on evidence

If you include a number in any client-facing artifact, name the source beside it. If you cannot name the source, remove the number. Qualitative patterns and anonymized narratives are enough to make the case for sequencing, audits, and executive briefings.

Closing

An AI citation audit is the entry ticket to serious AEO work. Run the sixty-second check before strategy calls. Expand to a logged, versioned, competitor-aware system monthly. Tie gaps to backlog. Never invent share-of-model numbers.

Need help installing this checklist across your book of business? Start a strategy conversation with SEO Outsourcing. More operator essays sit on the blog; methodology background on about.