There’s a particular kind of meeting that happens about three months into most AI visibility projects. Someone asks, “Is it working?” Everyone looks at each other. One person remembers that ChatGPT used to get the service area wrong. Another thinks Perplexity mentioned the company last week, or maybe that was a competitor. A third pulls up a screenshot with no date on it.
Nobody is lying. Nobody can prove anything, either.
The fix for that meeting happens on day one, before any strategy is written: a dated baseline. It’s a structured record of what machines see and say about you, captured before you change anything. It’s the least exciting deliverable in answer engine optimization and the one that makes every later deliverable credible.
What is a dated baseline in AI visibility work?
A dated baseline is a snapshot of your current state across the signals you plan to influence, with each observation stamped with the date it was taken and the method used to take it. It answers one question: what was true before we touched anything?
It isn’t a score you’re trying to beat. It’s a reference point. Without one, you can describe the present. With one, you can describe change.
Why have record-keepers always cared about the “before”?
Some of history’s most useful documents were unglamorous baselines.
The Domesday Book. Compiled in England in 1086, it recorded landholdings and resources across much of the country. Historians still use it because it captured a moment with enough structure that later records could be compared against it.
Weather records. National weather services keep long, dated observation series because a single reading means little on its own. Climate claims, for or against anything, depend on records that were kept consistently before anyone knew what they’d be used for.
The national census. The United States has conducted a census every ten years since 1790. Its value comes not from any one count, but from asking similar questions on a fixed schedule, so the population’s change becomes visible.
Your AI visibility baseline is a modest cousin of these. Same principle: record consistently, date everything, and resist the urge to tidy up what you find.
What should the baseline include?
Nine observations make a solid starting record for most small businesses. Group them in three sets.
Identity signals
- Canonical name usage: where your exact business name appears, and where variants show up.
- About page first paragraph: copy and paste the current text into your record.
- Organization schema and sameAs: whether structured data exists and which profiles it links to.
Access signals
- robots.txt contents: a full copy of the file as it stands today.
- Answer-ready pages: a list of pages that directly answer common customer questions.
- Freshness: last meaningful update date for your key pages.
Perception signals
- AI assistant answers: verbatim responses from two or three assistants to a fixed set of customer-style questions.
- Third-party mentions: a short list of where you’re referenced off-site.
- Search appearance: how your brand search results look, noted in plain words or captured as dated screenshots.
The Free AI-Visibility Quick Check covers these signal types in ten quick checks, and its final check is exactly this: a dated baseline. If you want a fast structure to fill in, start there.
How do you record AI assistant answers so they’re comparable later?
This is where most baselines fall apart, because AI answers vary. Ask the same question twice and you may get two different phrasings. That’s not a reason to skip the record. It’s a reason to make the method consistent.
- Fix the questions. Write five to ten customer-style prompts and reuse them word for word each time.
- Fix the context. Note whether you were signed in, which assistant and mode you used, and whether web browsing was on.
- Record verbatim. Paste the full answer. Summaries drift; quotes don’t.
- Classify, don’t grade. For each answer, note whether it’s correct, absent, confused (merged with another entity) or stale. Those four labels are more useful than a 1–10 score.
- Date everything. The date is the whole point.
Over time, you’re not looking for a single answer to change. You’re looking for the pattern across your fixed questions to shift.
What’s the most common baseline mistake?
Fixing things while you record them. You open the About page to copy the first paragraph, notice it’s terrible, and rewrite it on the spot. Now your baseline records the new version, and you’ve lost the “before.”
Discipline here is simple: record first, fix later. Finish the baseline in one sitting if you can. Keep a separate list of the fixes you’re itching to make. You’ll get to them, and the record will show the difference.
The second most common mistake is recording only what’s wrong. A baseline should capture what’s working, too. If an assistant describes you correctly today, that’s worth knowing, because a future change could break it.
How often should you re-baseline?
Treat the original as permanent and add new dated snapshots on a schedule. Quarterly works for most small businesses. Monthly makes sense during an active project. The key is consistency: same questions, same method, same structure, so each snapshot is comparable to the first.
Never overwrite the original. A baseline you edit is just a draft.
How does a baseline change the way you work?
Three things improve immediately.
Prioritization gets honest. When the record shows that two of three assistants misstate your location, that fix rises to the top on evidence, not instinct.
Reporting gets credible. For consultants, a dated before-and-after is the difference between “we improved your AI visibility” and “here’s what assistants said on the first date, and here’s what they say now.” Clients can read the change themselves.
Experiments get readable. If you change one thing between snapshots, you have a fair chance of connecting cause and effect. If you change six, you’ll know something happened. The baseline encourages smaller, cleaner batches of change.
Where should the baseline live?
Somewhere boring and durable. A spreadsheet works. So does a dated document in a shared folder. What doesn’t work is a pile of screenshots on someone’s desktop with filenames like “Screenshot (14).png.” The record has to survive staff changes, laptop replacements and the natural human tendency to tidy things up.
Three rules keep a baseline usable:
- One home. Pick a single location and link to it from wherever your team keeps project notes.
- Read-only original. Once the first snapshot is complete, lock it or duplicate it. Future snapshots go in new tabs or new files.
- Method notes on the same page. Write down exactly how you gathered each observation. A future colleague, or a future you, will need to repeat it.
What does a baseline look like for a consultant’s client?
For freelancers and agencies, the baseline doubles as a discovery deliverable. Run it in the first week of an engagement and hand the client a short summary: here’s how your business is described today, here’s what’s correct, here’s what’s missing, here’s what’s confused. Clients often learn more about their own visibility in that one document than in months of rank reports, because it’s written in the language they actually care about: what are people being told about us?
It also sets expectations cleanly. When the summary records that one assistant didn’t know the business existed on the first date, nobody later argues about whether that was always the case. The record settles it.
A useful habit is to let the client choose two of the fixed questions. They know which questions customers ask in real life. You know how to make those questions repeatable. Between you, the baseline measures what matters to the business, not just what’s easy to check.
What comes after the baseline?
Once the record exists, you can diagnose. A one-page check gives you the outline. A structured audit fills it in. The AI-Visibility (AEO) Audit includes an AI Prompt Log tab built for exactly this kind of repeatable spot-checking, plus 60 scored checks and a fix plan that ranks what to do first. It’s designed to be re-run on a schedule, which makes your baseline the first entry in an ongoing record rather than a one-off exercise.
What should you leave out of a baseline?
Opinions, mostly. The baseline is a record of observations, not a strategy memo, so keep interpretation in a separate notes column or a separate document. “Assistant B says we’re in Duluth; we’re in Superior” is an observation. “Assistant B is clearly pulling from our old directory listing” is a hypothesis. Both are useful. Only one belongs in the record.
Also leave out anything you can’t repeat. If you got a striking answer once from an unusual prompt, note it in the hypothesis file, but don’t add it to the fixed question set unless you’re prepared to ask it, word for word, every time you re-baseline.
Start the record today
The best time to have taken a baseline was before your last website change. The second-best time is this afternoon. It takes less time than the meeting where everyone tries to remember what things used to look like.
Download the Free AI-Visibility Quick Check, run it before you change anything, and put today’s date at the top. Everything you do next will be easier to explain.
The Quick Check, and the Audit that turns a baseline into a plan, are both on Content & SEO Workflows.
Leave a Reply