The AI Citation Scorecard: How Leaders Measure Whether AI Search Is Sending Revenue

Stop paying for AI marketing work that does not move one of three numbers: citation rate on your fixed prompt set, accuracy of those citations, or assisted pipeline that named search or AI. Everything else is optional.

What the scorecard actually decides

The AI citation scorecard is not a report. It is the decision tool. Keep one fixed list of roughly 25 prompts that match how buyers ask for your category. Run them monthly in ChatGPT, Perplexity, Google AI Overviews, and Gemini. Record three columns only:

• Cited or not.
• Accurate description of what you sell and where you operate.
• Assisted pipeline that named search or AI as the source.

If a month ends and none of those three moved, the work that was funded that month did not earn its keep. The conversation with the partner or the internal team is simple: keep, fix, or stop.

What to stop paying for

Most AI marketing retainers fail the same way. They produce activity that never touches the scorecard.

Stop paying for vanity mentions. A screenshot of your name in ChatGPT is not evidence. If the description is wrong, or the prompt was never on your buyer list, the mention has no commercial value. Demand the full prompt, the full answer, and the accuracy note.

Stop paying for tool stacks that do not feed the scorecard. Content calendars, automated schema generators, and “AI visibility platforms” that cannot show you the same 25 prompts month over month are overhead. If the tool cannot export the three columns above, cancel it.

Stop paying for retainers that refuse to kill tactics. A partner who will not identify what to stop after 90 days is protecting the invoice. Ask for the kill list in the first scope document. If it is missing, do not sign.

Stop paying for local work that treats a ranking drop like a suspension. A suspended Google Business Profile is removed. Recovery is an appeal. A map-pack ranking drop leaves the profile live but lower. The fixes are categories, review velocity, citation consistency, and on-page content. Never open a second profile to “replace” the first. That creates a new problem Google will also suspend.

How to measure pipeline from AI search

You do not need an attribution platform. Add one field to every form and one question to every sales call: “How did you find us?” Keep the options open enough that “ChatGPT,” “Perplexity,” “Google AI,” and “search” can appear. In the monthly review, count only the assisted conversions that named one of those sources.

Do not invent a percentage of total revenue. Report the raw count of forms, calls, and booked consultations that named search or AI, alongside the citation rate and accuracy from the scorecard. If the citation rate rises and the assisted count does not, the citations are not on prompts that create pipeline. Change the prompt set or the offer. If the assisted count rises while citation rate stays flat, you may be capturing branded search that AI answers are influencing. Either way, the three numbers tell you what to keep funding.

For the local side, the same discipline applies. If map-pack rankings dropped, review the Google Business Profile categories, recent reviews, and citation consistency before you approve more content or ads. If the profile was suspended, the only path is the official appeal. A second profile is not a recovery plan.

The 90-day decision

Week 4: baseline the scorecard and the assisted pipeline count. Week 8: correct the highest-volume inaccurate citations. Week 12: compare the three numbers to baseline. Then decide:

1. Keep the work that moved at least one number.
2. Fix the work that is close but incomplete (usually entity data or prompt coverage).
3. Stop the work that produced activity with no movement.

Write the decision in one sentence for the next monthly meeting. Operators who do this stop paying for the same non-working tactics two quarters in a row. Operators who do not keep funding reports that never change the business.

What to look for before you sign the next engagement

Ask three questions before the next invoice:

• Will you use our existing 25-prompt scorecard, or will you invent a new one?
• How will you show assisted pipeline, not just mentions?
• What will you recommend we stop paying for at day 90?

If the answers are vague, the work will be vague. The scorecard and the pipeline count are the only evidence that matters.

Specific steps for the next 30 days

Pull last month’s sales calls and form submissions. Count how many named ChatGPT, Perplexity, Google AI, or search. That is your assisted baseline. Do not estimate. If the field does not exist yet, add it this week and start the clock at zero.

Open your current AI or SEO report. Find every claim that does not map to citation rate, accuracy, or the assisted count above. Circle those claims. In the next meeting, ask which of them you will stop funding. Write the answer down.

If you have a Google Business Profile, check its status in the dashboard. Suspended is an appeal. Ranking lower in the map pack is categories, reviews, and citations. Record which one you have. Do not approve work that treats them as the same problem.

Finally, send the three questions above to any partner who will invoice you next month. The replies become the scope. If the replies do not mention the scorecard or the assisted count, do not renew.

Operators who run this once per quarter stop paying for the same non-working tactics twice. The alternative is another year of reports that never change a decision. The scorecard and the assisted pipeline count are the only numbers that should appear in the executive review. Everything else is supporting detail for the team that owns the work.

Frequently Asked Questions