You changed the page. How long before you judge whether AI answers moved?
Most AI visibility "tests" I see fail in the same place. Someone edits a page, asks ChatGPT the same question the next morning, sees no change, and decides the edit did nothing. Or the answer changes once and they call it a win.
Both conclusions come too early. Here is the waiting routine I use now.
-
Freeze the question set before you touch the page.
Write down five to ten buyer questions, word for word. Run them once and save the answers with the date and the tool name. That is your before state. If you change the wording later, you are testing a new question, not your page. -
Keep one wait per assistant.
The assistants do not pick up changes at the same pace. Copilot can look faster when the page is already in Bing. Tools that lean on their own crawl or on older training data can lag for weeks. So I don't use one global "check back in a week". I note a separate recheck date for each tool. -
Only count a move that repeats.
Answers vary between runs even when nothing changed. One flip means little. I only count a change when the same question gives the new answer on two runs a few days apart. -
Separate what moved.
Did the tool start naming you, start citing your page as a source, or start recommending you? Those are three different results and they move at different speeds. A new mention without a citation usually means the model knows the name but is not leaning on your page yet. -
Change one thing per round.
If you rewrite the intro, add an FAQ and fix robots.txt in the same week, you will never know which one did it.
This is slow and a bit boring. But it is the difference between a log you can trust and a screenshot that happened to look good that day.
What wait times are you seeing per assistant? I'm especially curious whether anyone has seen Perplexity pick up a change faster than Copilot.
Very useful routine. Saving the exact buyer questions and dated answers is the part most people skip. One addition: make sure the page is readable without JavaScript. A lot of "no change" results are just crawlers seeing an empty shell.
Good addition. A page that only renders with JavaScript can look like no change at all, when the crawler just saw an empty shell. I now check the raw HTML first, before I start any wait.
The “freeze the question set before you touch the page” step is the one I’d steal even outside AI visibility.
Same trap shows up when people track competitors: they rewrite the scoring criteria mid-week, then can’t tell whether the market moved or their rubric did. I’ve started dating the exact checklist next to the dated answer — and I won’t call something a move until it shows up twice with the same checklist.
Curious: when Perplexity does pick up faster for you, is it usually because the page was already in their citation graph, or because you also got an external mention that week?
Mostly the first. When Perplexity moved within days, the page was usually already being cited for a nearby question, so it was more of a refresh than a discovery. The few times a new page moved fast, there was also an outside mention that same week, so I can't cleanly separate the two yet. I now log outside mentions in the same dated sheet, so at least I can see when they overlap.
the part that bit me is the answer isn't always reading the page I just edited. I'll change the homepage sentence and the tool still quotes some old comparison page or a directory blurb from weeks ago. so the wait looks endless when the source that actually got cited never changed. I started writing down which URL the answer cites, not just whether my name showed up. if the citation is still the old page, the homepage edit hasn't been tested yet.
Yes, this is the step most people skip. If the answer still cites the old comparison page or a directory blurb, the page you edited hasn't been read yet, so the clock hasn't really started. I now note the cited URL every time and only start counting once the answer quotes the page I changed. Sometimes the quicker fix is to update that old source instead.
The “only count a move that repeats” rule makes a lot of sense. A single change can easily be noise, especially when different assistants update at different speeds. I’d be interested in seeing whether the lag is consistent for the same type of change, or whether it varies more by crawlability and source authority.
From what I've seen it varies more by the page than by the type of change. A page that is already cited elsewhere or sits well in Bing moves faster, while a new page with no outside mentions can sit for weeks even after a small edit. So I log the change type and the page's starting position together, otherwise the lag numbers are hard to compare.
That makes sense. If the starting authority and visibility of the page are different, comparing lag times without that context can be misleading. A small change on an already-established page may get picked up quickly, while a larger change on a low-visibility page might barely register for weeks. The baseline matters as much as the change itself.
Exactly. That's why I stopped comparing lag across different pages and only compare a page against its own earlier runs. It's slower to build up, but the numbers finally mean something.
Yeah, that makes much more sense. Comparing each page against its own baseline gives you a much cleaner signal and makes the results easier to trust
Nice, this makes a lot of sense. What's been the most surprising part of it so far?
Honestly, how often the answer flips when nothing changed. Running the same question twice a few days apart showed more noise than I expected, so a single before and after tells you very little.