The short answerDo not credit or blame a migration for a change in AI visibility unless you recorded a baseline before launch. Check indexing and bot access, then read each engine alone over weekly windows. In our 14-day study of one category, an agency's daily named rate on a question sat on average about 30 points from its rate on other days.
What to track, at a glance
| Measure | What it tells you | How to read it | The trap |
|---|---|---|---|
| Indexed pages | Whether Google can show a page as a supporting link in its AI features | Check every key URL on the new site in the first week | AI feature clicks stay inside Web search; the separate AI view shows impressions only |
| Bot access | Whether each engine's search bot can fetch your pages | Read robots.txt and your firewall rules, bot by bot | User-triggered fetchers may ignore robots.txt |
| Retrieved | A page entered the candidate set for an answer | Count it per engine over a window | On Perplexity, Peec found 64% of retrieved URLs were never cited |
| Cited | The answer used and clearly referenced the page | Compare it with retrieved for the same pages | An engine can cite your page and still not name you |
| Named | The answer text mentions your brand | The number buyers see | ChatGPT repeated its exact agency list the next day 2.9% of the time in our study |
| Visibility | The share of answers that mention your brand | Your headline reading, per engine | High visibility can sit beside a low share of voice |
| Share of voice | Your mentions against all brands mentioned | Read it against the same competitor set | It moves when a competitor moves, with no change on your side |
| Position | Where you appear in the answers that name you | Closer to 1 is better | Untracked brands count in the rank |
You cannot read a migration off any of these numbers. Your site, your content and the engines all change at once, and your job is to pull them apart. Work in order: baseline, site checks, readings, then the three-month watch. For the wider site checks, use our SEO migration checklist.
Set the baseline before launch
1. Fix the question set. Pick a fixed set of buyer questions, worded the way buyers ask an AI assistant, and do not edit them mid-way. On one migration we tracked, AI visibility rose in the months after the move, but the number of answers tracked each month fell by more than half and new content went live alongside it. The move gets no credit. If you cannot name your question set, you have no baseline.
2. Record every engine on its own. ChatGPT, Perplexity and Google AI Overviews at least. On the Google side, our advice is to export the generative AI performance report in Search Console before launch as well. In our study of 987 same-day answer sets, all three engines named a tracked agency in 500 sets, and in those 500, 71.3% of agency appearances came from only one engine and 7.8% appeared on all three. ChatGPT named at least one tracked agency in 86.5% of its answers, Google AI Overviews in 83.3% and Perplexity in 58.0%. The study ran one week, one run per engine per day, on B2B SaaS SEO, AEO and growth-agency questions in US English, and counts only the 53 brands Peec tracks in our project, almost all of them agencies. An average across engines hides which one moved, as our study of how far the engines disagree shows.
3. Write down the definition you use. Peec defines visibility as the percentage of AI responses that mention your brand, share of voice as your mentions compared with all brands mentioned, and position as the average rank where you appear. Whatever tracker you use, freeze three things before launch: the definitions, the questions and the engines. Change one after launch and you compare two different things.
4. Separate retrieved, cited and named. Peec defines retrieval as a page entering the candidate set for an answer, and citation as the model actually using and referencing the page. On Perplexity, Peec reports that 64% of retrieved URLs were never cited (February 2026 data). In our own study of one week of tracked answers about our category, 98.9% of the answers that named LoudFace had a loudface.co page among their sources. That study counts a page in an answer's sources as a read, covers one brand over one week, and found that engines named us almost only when they had read our page. It also found that being read was not enough: an engine can cite your page and still not name you.
5. Log what else ships. Write down the launch date and every piece of content that goes live in the weeks around it. Without that log, you cannot tell a move from a new article.
Check the site in week one
6. Make every key page indexed and eligible for a snippet. Google says a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link in AI Overviews or AI Mode. It says there are no additional technical requirements, and that indexing and serving are not guaranteed. Google also lists allowing crawling in robots.txt, and by any CDN or hosting infrastructure, among the fundamentals. A move often changes that infrastructure, so test it on the new site.
Your Google-side reading already exists. Google says traffic from AI features is counted in the Search Console Performance report under the Web search type, so a drop there mixes classic clicks with AI feature clicks. As of August 31, 2026, Search Console also has a generative AI performance report for all websites, showing impressions for AI Overviews and AI Mode by page, country, date and device. Its help page lists impressions only, with no clicks.
7. Keep the search bots allowed. Check each bot against your new robots.txt and firewall.
| Bot | What it does, per its owner | What blocking it does |
|---|---|---|
| OAI-SearchBot | Surfaces sites in ChatGPT's search features | The site is not shown in ChatGPT search answers, though it can still appear as a navigational link |
| GPTBot | Crawls content that may be used for training | Signals that the content should not be used for training |
| ChatGPT-User | Fetches pages for user actions in ChatGPT | robots.txt rules may not apply to it |
| Claude-SearchBot | Improves search result quality for users | Prevents Anthropic from indexing your content for search, which may reduce visibility and accuracy in user search results |
| Claude-User | Fetches pages for user requests in Claude | May reduce visibility in user-directed search |
| PerplexityBot | Surfaces and links sites in Perplexity search results | Perplexity recommends allowing it so your site appears in those results |
| Perplexity-User | Fetches a page when a user asks Perplexity a question | It generally ignores robots.txt, because a user triggered the fetch |
Each robots.txt setting is independent at OpenAI, so you can allow OAI-SearchBot and disallow GPTBot. Both OpenAI and Perplexity recommend allowing requests from their published IP ranges, and Perplexity says a site behind a web application firewall may need to explicitly allow its bots. A robots.txt change can take about 24 hours to register at OpenAI, and up to 24 hours at Perplexity. Put a page that must stay private behind a login, because robots.txt may not stop the user-triggered fetchers.
8. If your buyers use Microsoft 365 Copilot, check Bing indexing. Microsoft says Copilot and Copilot Chat in Microsoft 365, with web search enabled, may fetch information from the Bing search service to give a grounded response. That statement covers those products only. Confirm your key pages are in Bing's index after the move.
Read the numbers the right way
9. Read a week against a week. Our rule: judge a week against a week, and a trend over several weeks. Google says AI Overviews and AI Mode may use different models and techniques, so the responses and links they show will vary. And in our study of 1,636 answers over 14 days, the identical agency list came back the next day only 2.9% of the time on ChatGPT, 20.3% on Perplexity and 5.8% on Google AI Overviews. On average, an agency's named rate on a question on a single day sat 30.6 to 30.7 points from its rate on that question's other days, and its week-one rate sat 18.5 to 20.3 points from its week-two rate. That study counts only the 53 brands we track, in one category, from a logged-out tracker set to a US location, with one run per question per day. Read its method and limits before you reuse them. None of these studies measured a migration. Ask how many days stand behind each point on any chart you are shown.
10. Throw out the days that are not ordinary. Perplexity returned about 20 sources per answer on 12 to 15 September 2026 instead of its usual 10. That doubled citations and brand visibility for every brand we track for four days, so any window containing those days reads as a fall afterwards. Our advice: when an engine changes how it answers, leave the change days out and compare the weeks on either side.
11. Judge by the pages that enter answers. Google says AI Overviews and AI Mode may use a query fan-out technique, issuing multiple related searches across subtopics and data sources. The searches behind one question shift from run to run. In our study of 450 answers to 10 agency-buyer questions over 15 days, ChatGPT showed its search queries in 101 of 150 answers, and across 91 pairs of consecutive runs that showed queries, none matched. The study's own advice in its comparison of query fan-out tools is to read the patterns across runs, never single strings. Track which of your pages enter the answers, and leave the query strings out of your migration report.
Already migrated with no baseline
In our judgment, start now: pick your question set and engines today, call today day one, and judge only the trend from here forward. Do not rebuild a before picture from old screenshots, because you cannot recover the questions behind them. Google says the Search Performance report can show the last 16 months, with AI feature traffic counted under the Web search type, so use it for the Google side and read it as a blend of classic and AI clicks. Google says the usual Performance report data limits, such as the time period, also apply to the generative AI performance report. In our judgment, start recording that report now: export its impressions for AI Overviews and AI Mode today and keep the file with your day-one readings. If you changed domain, keep the old property: Google's Change of Address tool requires you to own both the old and new properties.
What normal looks like, and when to act
Google gives no ranking recovery time after a move. For small and medium sites it says most pages move over a few weeks, and larger sites take longer. The table below is our judgment. No study, ours included, has measured these windows after a migration.
| Window (LoudFace judgment) | Read this | A normal wobble | Act when | Then do this |
|---|---|---|---|---|
| Week one | crawl and index checks, bot access | large swings on any one engine | a search bot is blocked or a key page is not indexed | unblock the bot or fix the page that day |
| Weeks two to four | retrieved and cited pages per engine | names dropping in and out of single answers | your pages stop being retrieved on one engine for two weeks in a row | check the redirects and content of the pages that engine used to cite |
| Month two | visibility and share of voice per engine, week against week | an agency's named rate on one question moving about 20 points between weeks | one engine stays below your baseline for two weeks in a row | read which pages that engine cites in place of yours |
| Month three | the same readings against your baseline | share of voice moving with a competitor | you sit below baseline on every engine for two weeks in a row | treat it as a site or content problem and audit the key pages |
What never to promise
12. Never give a recovery date. Google gives none, and the engines change from day to day. Anyone who promises you a date is guessing.
13. Claim only what you measured. We have no clean reading of what a migration does to AI visibility, and one migration we tracked was mixed up with changes in the tracked questions and new content. A clean test needs a baseline taken before launch and a content calendar that stays still.
14. Say whose data it is. Every number in a migration report should carry its tool, engines, question set and dates. If a report cannot name those four, do not act on it.
When it is worth bringing in help
Bring in help before you move a site that earns AI citations today, so the baseline and tracking are running before launch day. Our AI visibility audit guide sets out the ten checks a real audit covers, from crawler access to visibility per engine and the pages each engine cites. Our guide to reading your server logs for AI-bot traffic shows how to find GPTBot, ClaudeBot, PerplexityBot and the rest in your own logs and check the traffic is real.
Book a free 30-minute intro call before the move. We pull up your live site and give you an honest read on what we would change and why.
