Measuring your AI visibility — a director's simple scorecard

The scorecard

The Monday-morning exercise from the start of this series — ask an assistant what a family would ask, and read what comes back — was meant to be done once, to find the gap. The scorecard is that same exercise given a shape you can repeat, so the gap becomes a number you watch move.

Five questions, asked the way your county asks them:

1. Who should I call when someone dies at home in [your town]?
2. Best funeral homes near [your town].
3. How much does cremation cost in [your county]?
4. Tell me about [your funeral home’s exact name].
5. Is [your funeral home] locally owned?

The first three are discovery questions — a family who does not yet know your name, deciding who to call. The last two are branded — a family who already has your name and is checking you out. Both belong on the scorecard, but they measure different things: the discovery questions ask whether the machines find you at all; the branded questions ask whether, once found, you are described the way you would describe yourself. The branded search has a deeper audit of its own — Your funeral home’s name is a query — audit what it returns walks the full results page — so here it is only two rows, a temperature check.

Put each question to two or three assistants — a general chat assistant, the AI-summarized answer at the top of a search, and one more if you use it — in a fresh session, not signed in, so your own history does not tilt the answer toward the home you already run. Read the paragraph that comes back. For each answer, record three things: were you named, what was said about you, and what was wrong.

That is the whole instrument. Ask the first question about [your town] and maybe the assistant names two funeral homes and yours is not among them — that is one row, and its mark is simple: not named. Ask the branded question and maybe it names you but calls you cremation-only when you also handle burials — named, but described wrongly. You are not writing an essay; you are making a handful of plain marks.

One lane has grown a real dashboard since this series began: Bing Webmaster Tools now runs an AI Performance report showing where AI-generated answers cite your site — Copilot and Bing’s own surfaces, in public preview since February 2026. Verify your site there and check it the same quarterly morning. But the answer a family actually hears, assistant by assistant, is still gathered the way this page gathers it: by asking.

Score it plainly

Three marks per question, one for each thing you recorded:

• Presence — were you named at all?
• Accuracy — was every fact right: the phone number, the services offered, who owns the home?
• Depth — did the answer carry a specific, a review, a story? Or just a generic line that would fit any funeral home in the county?

A point for each. Five questions, three marks, one date at the top: fifteen cells on a notepad or a single sheet. No spreadsheet heroics, and — this matters — no benchmark. There is no industry average to score against; anyone who sells you one is guessing, because these answers are not published anywhere to be averaged. The only honest comparison is you against your own last quarter.

That is where the value sits: the direction, not the decimal. Last quarter you were absent from three of the five answers; this quarter, from one. That sentence is the whole point of the exercise. It does not need a percentage to be true, and a percentage would only invent a precision the underlying thing does not have — ask the same question twice on the same morning and the wording will shift, because the models compose each answer fresh. That is measured, not folk caution: when SE Ranking put 10,000 queries through Google’s AI answers three times in a single day, 91% of the cited pages changed between identical runs. Do not over-read a single run. Ask each question, take the mark that holds across a couple of tries, and trust the quarter-over-quarter movement rather than any one paragraph.

Close the loop

The scorecard earns its fifteen minutes because every mark on it points at a fix already written down in this series. It is not another job; it is the ordering of the jobs you already have.

• Absent — the machines are not finding you on the discovery questions. That is the work of pages that answer real questions in plain words (Write the page for the question they actually ask) and the steady pulse of obituaries on your own site, which prove the home is active this week.
• Named but wrong — the facts exist but disagree across your site, your Google Business Profile, and the directories, and the machine repeated the wrong one. That is the consistency audit: make the name, address, phone, hours, and services identical everywhere, so there is no wrong version left to quote.
• Named but thin — you are in the answer, but described in a line that would fit anyone. That is the work of reviews that tell stories and of your own specifics in writing (Reviews that answer questions beat reviews that award stars) — the details a machine can lift to make the mention yours.

One more distinction the scorecard should hold: the three-pin map box and the one-paragraph AI answer are different games, and a funeral home can sit in one and vanish from the other. Keep them on separate rows when they diverge — the fix for a missing map pin is not the fix for a missing paragraph.

Read this way, a quarter’s sheet is a to-do list you did not have to invent. The lowest column is next quarter’s project.

Why quarterly is enough

Checked weekly, the scorecard measures noise — the day-to-day flicker of models composing fresh answers, not any real change in how you are seen. Checked yearly, it misses the things that actually move: a competitor’s new campaign, a directory that started publishing a wrong address, a run of reviews that reshaped your description. The substrate underneath — the models, the directories, your own obituary cadence and review flow — turns over on the scale of months. Quarterly is the rhythm that matches it.

So put it on the calendar. Four dates a year, fifteen minutes each, next to the tax deadline you never miss. The score itself is almost beside the point; the habit is the asset. A funeral home that looks, four times a year, at how the machines describe it — and fixes what it finds — is running a marketing operation most homes in your state cannot claim, for the cost of an hour a year and the willingness to read what comes back.