GEO reporting for management: the one-page report, and the decisions it should ask for

Who the page is for, and what they need from it

The reader is the owner, a partner or a manager who pays for the work and has a few minutes for it. They will not read answer logs. They want to know whether the assistants are putting the business in front of buyers, whether anything the assistants say could cost a customer, and whether anything is waiting on them. A report that makes them work those three things out from a dashboard has handed them the analyst's job.

That sets the order. The verdict comes first, because it is what they came for. The numbers come second, because a verdict without them is an opinion. The decisions come last, because that is where the page asks for something, and the reader should arrive there already knowing why.

The page, top to bottom

Eight blocks sounds like a lot for one page, but most are two or three lines. The rates are a small grid, the competitors a list of five names with counts, the site facts four ticks. A block that runs longer than a few lines belongs in the appendix, with a pointer from the page. What each number means and how it is counted is set out in the seven GEO KPIs worth reporting; this page is about putting them in front of someone who has to act.

Write the verdict last, and put it first

The verdict is the hardest part to write and the only part some readers will read. It says what happened in words the reader would use, says where it happened, and claims no more than the numbers show. Draft it once the numbers are in, then move it to the top.

• A good one: "Named more often on problem questions this month, by Claude and Perplexity; still not named by Gemini. One wrong opening time found and corrected." It gives a direction, a place and a fact the reader can check.
• A weak one: "Visibility score up six points." It says nothing about where, and six points may sit inside the range between runs.
• A wrong one: "AI now recommends us." It turns a monthly rate into a promise, and next month's runs may not repeat it.

Show the range, so a small move is not read as news

The assistants answer the same question differently from one run to the next. Our methodology handles that by asking each question once per reading, dating the reading, and reading the pattern across twelve questions, and with monitoring across months, rather than any one answer. Even so, a change of a few points between months can be nothing at all. If your panel has several runs a month, mark the range on the trend line, the lowest and the highest of the month's runs, so the reader can see at a glance when a move is bigger than the variation.

Two rules keep the page honest. Call a change real only when it holds for two months, or shows on more than one assistant at once. And when every rate falls together on every assistant, look at access before anything else, because the site is the one thing all four share. In our audit of 9,483 small business websites, 7.2% of the sites answered our crawler with a challenge page rather than the site, and a firewall or plugin change can put a site in that group overnight. The crawler access check reads what robots.txt asks of each assistant's crawler; whether a challenge page stands in front of the site takes a fetch of the homepage from outside your own network.

Ask for decisions, not approval

A report that ends with "any questions?" has left its most useful part out. Much of what keeps a small business out of the answers is a decision only someone with authority can make, and the report is the place to put each one to them, with the evidence from the page above it.

• Settle one name, one phone number and one address. In the audit, 43.1% of sites were inconsistent about who and where they are, and 36.9% showed more than one phone number. The fix needs the logins to every listing and a decision about which number is the number.
• Publish a price range. When the panel's cost question names a competitor who shows a price and you do not, the fix is a range on your page, and only the owner can agree what it says.
• Put reviews on the site as text. 64.7% of the audited sites showed a crawler no readable review signal. Choosing which reviews to quote, and asking customers for more, is the owner's call.
• Approve a page for a question you are losing. Each comes with the question, the names the assistants gave instead, and what those businesses' pages had that yours did not.
• Lift a crawler block. If a hosting provider or an IT policy is turning the assistants' crawlers away, the page can show it; changing it is a decision for whoever owns that risk.

Give each decision an owner and a date. Next month's page reports whether it was made, which keeps the list short and makes a stalled item visible without anyone having to chase it.

What goes in the appendix

The appendix is for the person who doubts the page, and it should let them check any verdict on it. It holds the question panel as written, every answer verbatim with its date and assistant, the per-run table the rates were computed from, and each statement marked wrong with the passage it came from. None of it goes on the page. All of it has to exist, because a number nobody can trace to an answer gets argued about instead of acted on. Setting up prompt monitoring describes the record the appendix is drawn from.

Rhythm, owner, and a panel that does not change

Send the page on the same day each month, from the same person, on the same panel. The panel is frozen once the work starts, because a changed question starts a new series and the trend line stops meaning anything. When the business adds a service, add its questions as a new line on the page and mark where the line begins, rather than folding them into the old totals.

Once a quarter, lay the last three pages side by side. That is where a trend becomes visible, where the effect of a decision made in one month can be looked for in the months after, and where the reader can judge whether the work is worth continuing. The monthly page is for acting on; the quarterly look is for judging.