Build Your Quarterly GEO Update for the Board with Qwairy’s MCP

Your deck has one row for AI search, and this builds it. The analysis is a metric you already hold; the work is making it survive the first question about how it was calculated, which makes this a communication exercise rather than an analytical one. If you need tracked AI-referred sessions and revenue instead of one auditable row, the ROI dashboard playbook reads those.

The GEO problem it solves

The failure is rarely the measurement. It is a large percentage arriving with no denominator attached, so the first person who asks "out of what" ends the discussion.

The job is definitional. Pick one field, state its numerator, denominator, prompt count, provider count and dates, then reproduce it next quarter. A number a board can audit beats a bigger one it cannot.

How the workflow breaks down

1. Take one metric with its denominator. get_brand_performance takes explicit dates and returns mention rate, source rate and share of voice with numerator and denominator, plus sentiment, prompt count, provider count and response total. Mention rate counts responses mentioning you or a monitored competitor, not every response. The payload has no composite or overall score.

2. Get the delta from the same field. Call it again with last quarter's dates: two windows of one metric is the only comparison keeping one denominator on both sides.

3. Add the competitive line. get_competitor_comparison returns your market rank and mention counts against DIRECT competitors. It is point-in-time: where you sit today, never a gain or a loss, and the rank is null when your row is missing.

4. Keep any trend chart separate. get_visibility_trend returns daily points and a direction of up, down or stable, calling anything within two points stable. Its rate covers every completed response, a different denominator from step 1.

5. Write the two sentences, then the scenario. No tool drafts the narrative. get_content_opportunities and get_provider_breakdown show the gap you extend forward, formula and assumptions written out.

What a run actually returns

A real window returned mention rate 1.08 percent as 10 of 928 relevant responses, and source rate 3.49 percent as 54 of 1548 citable ones. Sentiment came back as 60.5 with nothing behind it.

The row that survives a board is "AI Search Presence 1.08 percent (10 of 928 monitored responses, 155 questions, 5 providers)". The row that does not is the same figure alone, because the first question is always how it was measured and the second is why it differs from the other percentage on the slide. Sentiment stays off the row: a number with no denominator answers neither.

What to do with the output

Nothing here attributes a deal to an answer. The row's job is to be defensible.

One row, one field: the mention rate from step 1, printed as the rate, then numerator over denominator, with prompt count, provider count and dates footnoted. Methodology and any trend chart go to the appendix.

Keep the definition with the value: field, numerator, denominator, prompt and provider count, both date ranges. Next quarter's row is read against it, so disclose any measurement change.

Then the rule deciding whether the row carries a delta. A denominator of zero comes back as a rate of zero, a different claim: print no qualifying responses instead. State a direction only when both windows exist and their denominators sit within twenty percent of each other. That threshold is this playbook's convention: below it the delta measures your monitoring setup as much as your visibility, so print the current window alone.

The Complete Prompt

I am preparing a quarterly board update and need one board-ready row on AI search. If I monitor more than one brand, list them and ask which to use.

1. Pull brand performance twice with explicit dates: the last completed quarter and the one before it. For each window give the mention rate with its numerator and denominator, plus prompt count, provider count, response total and dates. Never call a figure a composite or overall score; the payload has none.
2. Build the row from the mention rate alone: the rate, numerator over denominator, then a footnote with prompt count, provider count and dates. If a window returns a denominator of zero, print no qualifying responses rather than a rate of zero. Compare the two windows only when both denominators exist and are within twenty percent of each other; otherwise print the current window alone, say the earlier one is not comparable, and give me both denominators.
3. Pull the competitor comparison for the current window and write one line placing us against the DIRECT competitors returned. It is point-in-time, so never call it a gain, a loss, or a competitor's tactic. If my own row is null, drop the line and say my brand had no mention.
4. Only if I ask for a trend chart: put the visibility trend in the appendix as its own measure and state its denominator, since it does not cover the same responses as the row. It calls anything within two points stable and still says stable with no data points, so check the points first and write no data rather than flat.
5. Draft two sentences: the measured state with the competitive frame, then the commitment for next quarter. Keep them apart so a promise never reads as a measurement.
6. Pull content opportunities and provider breakdown, then describe the gap if the current mix holds, listing the formula and every assumption. Label it a scenario, keep it in speaker notes, and never turn it into revenue.

Sentence two needs what the data cannot tell you: who owns the commitment, the review date, and my footnote conventions. Ask for those first.