Key Takeaways
- Fix the question list before you record anything. The same questions, word for word, every month. Change the wording and you are measuring your wording rather than your visibility.
- Record who was named instead of you. The list of other names is the most useful column in the log, because it tells you what the answer is being built from.
- Use a fresh chat every single time. An assistant that remembers you from earlier in the conversation will name you, and that result means nothing.
- Run it monthly, not weekly. Answers wobble on their own. Checking too often turns normal variation into a story you then act on for no reason.
- Read the trend across three runs, never one. One good month is noise. The same movement in the same direction across three months is a signal worth acting on.
You do not need a subscription to know whether AI is naming you. You need a fixed list of questions, a spreadsheet, and forty minutes once a month.
The reason most people never find out is not cost. It is that they ask slightly different questions each time, on no particular schedule, and end up with a pile of impressions instead of a record.
This is the routine that turns it into a record, and the honest account of what a hand-kept log can and cannot tell you.
How do you track AI visibility without software?
Keep one spreadsheet. One row per question, one column per month, and the same questions asked in a fresh chat on roughly the same day each month. That is the whole method.
Two rules make it worth keeping. The questions never change wording, and every run happens in a fresh chat with no memory of you. Break either one and you are measuring the test rather than the market.
The output is not a score. It is a direction: more of your questions naming you, fewer competitors dominating, or nothing moving at all. All three of those are useful findings.
The five things to record, and nothing else
Date, question, whether you were named, who was named instead, and one line on how the answer described your kind of business. Five columns. Anything more and the log stops getting filled in.
The third column is the obvious one and the fourth is the valuable one. The names that come back instead of yours tell you which pages the answer is being assembled from, and that is a reading list rather than a defeat.
Keep the wording of each question exactly as a buyer would type it. If you have not built the list yet, the starting set and the method for choosing questions are in the fifteen-minute test.
The monthly routine, start to finish
Same day each month, fresh chat, ten questions, one pass. It takes about forty minutes at ten questions, and the second month is faster than the first because the sheet already exists.
In text, because the figure holds no claim of its own, and because these numbers are illustrative. In the example, five questions are asked three months running. One question moves from not named to named after a follow-up. One moves all the way to named in the first answer. One was already improving and holds. One never moves at all, and one gets stuck halfway. That mix is what a real sheet tends to look like: partial progress, one stubborn question, and nothing dramatic.
The routine itself is six steps. Open a fresh chat with memory off. Ask question one and paste the answer into the sheet. Mark named, named after a follow-up, or not named. Write down the other businesses that appeared. Repeat for every question on the list. Close the chat and do not go back to it until next month.
What each column actually tells you
Five columns, five different jobs. Knowing which question each one answers is what stops the log becoming a chore with no payoff.
| Column | What you write | Why it is there | What a change in it means |
|---|---|---|---|
| Date of run | The day you asked | Keeps the interval honest | Gaps explain odd results before you blame anything else |
| Question | The buyer's exact wording | Fixed wording is what makes runs comparable | If this changes, the row starts over |
| Named? | Yes, after a follow-up, or no | The headline result | Movement here is the thing you are tracking |
| Who was named instead | Every other business in the answer | Shows what the answer is built from | New names mean the sources shifted, not just the ranking |
| How it described you | One line, in the answer's own words | Catches wrong or stale descriptions | A description improving before a mention appears is early progress |
The last row is the one people skip and then wish they had kept. A business often gets described correctly before it gets recommended, so watching the description change is the earliest signal available to you. On the cadence question, how often to check covers when monthly is wrong and something else is right.
How to read the log after three months
Look for the same movement in the same direction across three runs, and ignore everything else. A single month that looks great is not evidence, and acting on it is how people end up rewriting a page that was working.
There is a real reason for the wobble. These systems are not deterministic, and the material they draw on changes underneath them. Google describes how its AI features use web content, OpenAI publishes which crawlers it operates and what they do, and neither describes a stable ranking you can watch tick up. The Princeton study on optimizing content for generative engines measured changes in how often sources were used at all, which is closer to what your log is really recording.
Now the limits, honestly. A hand-kept log covers one assistant, a handful of questions, and one run each. It cannot tell you how often real buyers ask those questions, it cannot separate a change you caused from a change in the model, and at ten questions the sample is small enough that two rows moving is not a trend. That is the trade you accept for spending nothing, and it is usually a fair one.
If you reach the point where you need more questions, more assistants, or a record someone else can read, that is when tooling or a service starts earning its keep, and the honest comparison of monitoring tools and done-for-you services lays out what each actually covers. the spreadsheet is not a poor substitute. It is the same measurement, done by hand.
Our free check asks twenty-five buyer questions for you at $0 and emails a plain-English readout within the hour. It runs as soon as you ask for it.
Questions we hear the most
How do I track my AI visibility month over month without buying software?
Keep a spreadsheet with one row per question and one column per month. Ask the same questions in a fresh chat on the same day each month, record whether you were named and who was named instead, and compare across runs.
How many questions should I track?
Ten is enough to see a pattern and small enough that you will keep doing it. Twenty-five is better coverage if you have the patience for it, which is the number our own check uses.
Why does the same question give me different answers each time?
Because these systems are not deterministic and their sources change. Some variation between runs is normal, which is why one result is never worth acting on and three consistent runs usually are.
Do I need to check every AI assistant?
Not to start. Pick the one your buyers most likely use and track it properly. Adding a second assistant doubles the work and mostly confirms what the first one told you.
Should I use incognito or a logged-out window?
Yes, or a fresh chat with memory and personalization turned off. Otherwise the assistant may be answering with what it knows about you rather than what it knows about your market.
When is a free manual log not enough?
When you need many questions across several assistants, or a record you can hand to someone else. A spreadsheet cannot cover five tools at scale, and pretending otherwise wastes months.
How is this different from what the AnswerHalo free check does?
The same idea, run for you at a size you would not do by hand. We ask ChatGPT 25 buyer questions from your niche and send back the counts and the names, so your first month of data arrives without you building the sheet.
What does the AnswerHalo free check include?
It costs $0. We ask ChatGPT 25 real buyer questions from your niche, then email a plain-English readout within the hour: how many answers named you, who was named instead, and the first fix. It runs as soon as you ask for it.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →