Playbook ยท Foundational
How to Measure Your Visibility Without Fooling Yourself
A step-by-step playbook for choosing the numbers that actually tell you whether your search and AI-answer visibility is working, and refusing the ones that only look good.
Owners and marketers who are tracking rankings, impressions, or a single AI-answer screenshot and want to know if any of it is connected to booked jobs and revenue.
A small measurement setup: the outcomes worth watching, a repeatable way to read AI-answer visibility, and a habit of stating the method next to every number you report.
It is easy to build a dashboard that always looks good. Impressions climb, a follower count ticks up, one search term holds the top spot, and none of it tells you whether the phone rang or a job got booked. Meanwhile the harder questions, whether your ad spend caused a sale or merely rode along with one, and whether you are actually named in the answers AI engines give, get skipped because they are less flattering to check. This guide is about building a measurement habit that survives contact with a bad month: pick outcomes that matter, measure AI visibility as a rate instead of a screenshot, tell correlation from cause, and never let a number travel without the method that produced it.
Before you start
- At least one real business outcome you can count: booked jobs, calls, quote requests, or revenue.
- Access to whatever already tracks that outcome (a booking system, a call log, a point-of-sale report, or even a notebook by the phone).
- A willingness to write down a number you do not like.
What you need
- A spreadsheet.
- Google Search Console, if you have a website (free).
- Your booking, call-tracking, or point-of-sale system, whatever already exists.
- The AI engines themselves, ChatGPT, Google AI Overviews, Perplexity and Gemini, which you will use to check your own visibility.
The playbook
7 steps, in the order that pays off first.
-
Pick the outcome that pays your bills, not the one that flatters you
Impressions, follower counts, and a single keyword ranking are vanity metrics: real numbers that are easy to move in a flattering direction but that say nothing on their own about whether the business is better off. Before you track anything else, name the outcome that actually matters, booked jobs, calls that convert, quote requests, or revenue, and make that the number every other metric has to earn its way back to. If a metric cannot be connected to that outcome in one or two logical steps, it belongs in a footnote, not a headline.
- Name one primary outcome metric (booked jobs, calls, revenue) that every report has to tie back to.
- List the vanity metrics you currently report, and decide which ones move to a footnote.
- Confirm you have a real, current way to count the primary outcome, not an estimate.
Watch out A rising ranking or impression count with a flat outcome number is not progress. It is a sign the wrong thing is being tracked.
-
Measure AI visibility as a rate across repeated runs, never a single screenshot
The answer an AI engine gives to the same question can change from one run to the next, so a single good answer proves nothing and a single bad one is not failure either. Share of answer is how often you are named across a fixed panel of real buyer questions, checked more than once, with the engine and the date recorded next to the result. Build a small panel of the questions your buyers actually ask, run it on a schedule, and log the pattern instead of the highlight.
- Build a panel of 10 to 15 real buyer questions, in their words, not industry jargon.
- Run the panel across the engines you care about (ChatGPT, Google AI Overviews, Perplexity, Gemini) more than once, on more than one day.
- Log each result with the date and the engine, and calculate the rate at which you are named, not just a yes or no for the best run.
Watch out Never present one flattering answer, saved as a screenshot, as proof of AI visibility. Screenshots are a snapshot of a moving target, and a single one can be cherry-picked without anyone intending to lie.
-
Tell correlation from cause before you credit anything
Two numbers moving together does not mean one caused the other. A ranking improves the same month you also fixed your site speed, added reviews, and posted more often, and you do not actually know which change did the work, or whether the market simply picked up. Correlation is a starting point for a hypothesis, not a conclusion you get to report. Before you credit a tactic with a result, ask what else changed at the same time, and whether you would expect the same result if you had done nothing.
- List everything else that changed in the same window before crediting one tactic with a result.
- Ask whether a comparable period with no changes would plausibly show similar movement.
- Downgrade a claim from "this caused that" to "this happened alongside that" unless you tested it properly.
Watch out A confident causal claim built only on timing is the single most common way honest people end up reporting a false result.
-
Stop trusting last-click credit for what ads actually did
Last-click attribution hands full credit for a sale to whichever ad or link happened to be clicked last, even when that click came from someone who had already decided to buy. It counts presence, not persuasion, which is why platform-reported return on ad spend is almost always higher than the truth. Incrementality corrects for it: compare a group that saw the ad to a comparable group that did not, so the difference between them, not the total attributed to the ad, is what you credit. A simple holdout, a region or a segment you deliberately do not advertise to for a period, is the most accessible version of this most businesses can run.
- Identify one channel where you currently trust the platform-reported return on ad spend at face value.
- Set up a basic holdout: a comparable region, segment, or time window with no ad exposure to compare against.
- Calculate incremental ROAS (the return from the difference between exposed and holdout groups), and expect it to read lower than the platform number.
Watch out Every major platform is financially motivated to show you a flattering attributed number. Treat "our reported ROAS" as a claim to test, not a fact to repeat.
-
Anchor on a blended number that cannot be gamed channel by channel
Marketing Efficiency Ratio, total revenue divided by total marketing spend, is worth tracking precisely because it assigns no credit to any single channel, so shifting attribution between search, social, and ads cannot inflate it. When every channel report claims outsized credit but MER stays flat, that is the tell that you are looking at reporting artifacts, not real gains. Use MER as the anchor number in any monthly review, and treat any channel-level metric that contradicts it as the one that needs explaining.
- Calculate MER (total revenue divided by total marketing spend) for your last three to six months.
- Compare MER movement against the sum of what individual channels are claiming credit for.
- Flag and investigate any month where channel claims rise but MER does not.
Watch out MER is a health check, not a diagnosis. It tells you something is off; it does not tell you which channel is the problem on its own.
-
Never let a number travel without its method
A percentage, a rank, or a rate is worthless on its own, because the same word can describe a fair measurement or a cherry-picked one, and the reader has no way to tell the difference without the method stated beside it. Get in the habit of pairing every number you report with how it was produced: the date range, the sample, and what it is being compared to. This single habit, applied consistently, does more to keep a measurement practice honest than any individual technique in this guide.
- For every number in a report, write the date range and comparison point directly beside it.
- State the sample or panel size for any rate (share of answer, conversion rate, and so on).
- Reject any number, including your own, that arrives without a stated method.
Watch out A naked percentage with no baseline ("conversions up 40%") is not a lie by itself, but it is not a fact either. Treat it as an unfinished sentence until the method is attached.
-
Set a baseline before you change anything
You cannot measure the effect of a change you made after the fact, because you have nothing to compare it to except memory. Before you touch your website, your ad spend, or your content, record where every metric in this guide stands: your primary outcome, your share of answer, your MER, your current attribution assumptions. That snapshot is what every future claim of progress has to be measured against, and it is the single cheapest insurance against fooling yourself later.
- Record your current primary outcome number, share of answer rate, and MER before making changes.
- Date the baseline and store it somewhere you will actually find it again in three months.
- Set a fixed re-check date (monthly or quarterly) rather than checking only when a change looks good.
Watch out A baseline recorded after you already suspect the result is not a baseline, it is a rationalization. Set it before you act, not after.
Questions
Straight answers.
- What is the one number I should trust most?
- Whichever one is closest to money actually changing hands, booked jobs, calls that convert, or revenue, measured against a stated baseline. Everything else in this guide exists to help you interpret that number, not to replace it.
- Is share of answer the same as a search ranking?
- No. A ranking is your position in a list of links for a single search. Share of answer is how often you are named across repeated runs of real buyer questions on AI engines, which vary between runs in a way rankings do not. Treat them as related but separate readings.
- My ad platform shows a strong ROAS. Why would I not trust it?
- Platform-reported ROAS is built on last-click attribution, which credits the ad for sales that may well have happened anyway, a structurally optimistic measurement method. An incrementality test, even a simple holdout, gives you a more conservative, more accurate number.
- How often should I check my visibility?
- Enough to see a pattern, not enough to start peeking at a moving number and reacting to noise. A monthly cadence works for most small businesses: run your AI-answer panel, check your primary outcome, and calculate MER on the same schedule every time, so the comparisons are fair.
Want a measured read before you set your baseline?
A Machine-Readiness Score reads how findable your business already is across search and AI answers, with the method stated plainly, so your first measurement is a real one. It is a specialist-reviewed read, with no guaranteed outcome and no obligation.