You have a launch date. On that date you can lose every AI citation your business has, and not find out for three months. Redirects protect your rankings. They do not protect the confidence a language model has built in your URLs, and nothing in a standard migration checklist tests for it.
This page is about protecting that through a replatform, a redesign, a domain change or a CMS move. It is short on reassurance because the failure mode is quiet and the window to prevent it closes on your launch date.
What Actually Happens On Launch Day
A model cites a URL because it has accumulated confidence in it. That confidence comes from the page (its structure, its structured data, whether an answer can be lifted cleanly from it), from the entity behind the page, and from third-party sources that point at that specific URL.
A migration can break all three at once:
The URL changes. Every third-party source citing the old address now points at a redirect. Search engines follow it happily. A model that had learned “this URL answers this question” has to re-learn it, on its own crawl schedule, which may be weeks away.
The structured data does not survive the CMS change. This is the single most common one. Schema that was hand-added to the old templates does not get rebuilt in the new ones, and nobody notices because nothing visible changes. The page still looks right to a person and has become significantly harder to parse for a machine.
The content structure changes. A redesign that moves the answer below a hero section, a testimonial band and an intro paragraph has removed the extractable unit even though every word is still on the page.
The rendering changes. Moving to a JavaScript framework where main content renders client-side is the most severe version. Some crawlers will handle it. Several of the ones that feed AI systems will not, and will see an empty page.
Crawler access changes. A staging robots.txt with Disallow: / shipped to production is a
cliche because it keeps happening. A new WAF or CDN rule blocking unfamiliar user agents is the
quieter version, and it is invisible in every tool you already use.
Why You Will Not Notice For Months
This is the part that makes migration the highest-risk moment in a GEO programme.
Ranking loss is loud. It is tracked daily, it shows up in a tool someone looks at, and it is attributed to the launch immediately because the timing is obvious.
Citation loss is silent. Most businesses do not track AI citations at all, so there is no dashboard to go red. Each platform re-crawls on its own schedule, so the loss arrives in stages rather than as a cliff. Organic traffic often holds up, because the redirects are doing their job, which actively reassures everyone that the migration went well.
The pattern we see is: launch, six to ten weeks of everything looking fine, then a slow decline in referral traffic from assistants that nobody connects to a launch a quarter earlier. By the time somebody asks the question, the trail is cold and the third-party sources have long since settled on pointing at a redirect.
The Pre-Launch Baseline
Everything downstream depends on this, and it can only be done before you go live. Once the old site is gone, the baseline is gone with it.
We capture, on a fixed prompt set across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews:
- Which of your URLs are currently cited, for which questions, on which platforms
- Which competitors are named alongside you, and which are named instead of you
- The third-party sources that cite your specific URLs, so they can be found and updated afterwards
- Your current structured data, page by page, as a diff target for the new build
- Which pages carry the answer-first structure that is earning the citation
Screenshotted and dated, because a baseline you cannot show somebody later is not a baseline. This is also the only way to prove afterwards that the migration did or did not cost you anything, which matters if there is a supplier relationship to hold to account.
The Staging Review, Before The Build Is Locked
The cheapest possible moment to fix any of this is while it is still on staging. We work from a staging URL under NDA, with basic auth, alongside whoever is building it.
What gets checked: URL mapping against the cited-URL list from the baseline, so the pages that are
actually earning citations get priority treatment rather than being handled as generic redirects.
Structured data parity, type by type. Whether main content renders without JavaScript. Whether the
answer-first structure survived the template redesign. Crawler access on the staging config, so a
blocking rule is found before it ships rather than after. Entity signals, including whether the
Organization schema, canonical naming and sameAs links made the move intact.
The output is a list the development team can action, not a report. If your agency is doing the build, this is the point at which a specialist reading their staging site is worth most and costs least.
Launch Week
The first seven days decide how much of the next three months is spent recovering.
We monitor crawler access from the production logs (which is the only place a blocked crawler actually shows up), re-run the baseline prompt set to catch immediate drops, verify structured data in production rather than in staging, and confirm the redirect chain resolves in one hop for the URLs that were carrying citations.
Most of what goes wrong in launch week is fixable in launch week. Almost none of it is visible in Google Analytics.
The 30 And 90 Day Checks
Re-run the same prompt set, on the same platforms, against the same baseline. Thirty days catches the fast-moving platforms. Ninety days catches the slow ones and tells you whether third-party sources have started pointing at the new URLs or are still sitting on redirects.
Where a citation has not returned, that is the point at which we go after the specific third-party sources that need updating. That work is unglamorous and it is the reason the citation comes back.
What It Costs, And What A Lost Citation Costs
| Pre-launch baseline and staging review | Scoped as a Synaptic Audit, £2,950 one-off, credited in full against your first retainer month if you continue within 30 days |
| Launch week monitoring, plus the 30 and 90 day checks | Inside a retainer, from £2,950 a month on Foundation |
| Recovery after an unprotected migration | Longer, more expensive, and less certain, because part of it depends on third-party sources re-crawling on their own schedule |
The honest framing on value: nobody can tell you what one citation is worth, and you should be sceptical of anyone who puts a number on it. What is knowable is the asymmetry. Protection is a defined piece of work with a known cost, done once, on a date you already control. Recovery is open-ended work with an uncertain outcome, started at the point where you have already lost something and have to rebuild third-party corroboration you previously had for free.
That asymmetry is the whole argument. It is the same reason you back up before you migrate a database.
How Late Is Too Late
There is no point at which this becomes pointless, but the value curve is steep.
| When you call | What is still possible |
|---|---|
| Before URL sign-off, usually 4 to 6 weeks out | Everything. The decisions that matter most are still free to change. |
| Staging up, content not locked | Nearly everything. Structured data, rendering and structure are all still cheap to fix. |
| Launch week | Reactive but genuinely useful. Catching a blocked crawler or stripped schema in the first days avoids months of decay. |
| Post-launch | Recovery rather than protection. Slower, costlier, and partly dependent on other people’s crawl schedules. |
| Three months post-launch | Still worth doing, but you are now rebuilding rather than preserving, and there is no baseline to prove what was lost. |
If you are reading this the week before launch, that is still a useful week. If you are reading it the week after, the 30 day check has not happened yet and the trail is warm.
If You Are The Agency, Not The Brand
A good proportion of the people who need this page are the development, design or SEO agency doing the replatform, not the client whose site it is.
That is a reasonable position to be in. AI visibility protection is not a standard part of a migration scope, it was not in the estimate, and it needs a capability that has no reason to sit inside a build team. It is also the kind of thing that becomes your problem retrospectively if the client’s assistant referrals fall off a quarter after your launch.
We work alongside build teams routinely, and it can run white labelled under your brand if you would rather the client saw one supplier. Either way, the staging review is designed to produce a task list your developers can action, not a document that criticises their build.
Proof
MarGen migrated its own site. margen.net moved off WordPress onto a static build, with a full URL and template change, and then went from a near-zero baseline to over 295,000 search impressions in 90 days with referral traffic arriving from all five major AI engines.
That is a rebuild rather than a like-for-like replatform, so it is proof of the method rather than proof that your specific migration will be painless. It is also the reason the checklist above is specific: most of it comes from work done on our own launch, where the cost of getting it wrong landed on us.
Migration And AI Citations: Common Questions
Can a site migration lose you AI citations?
Yes, and more easily than it loses rankings. AI systems cite a URL because they have built confidence in it over time, drawing on the page itself and on third-party sources that point at it. A migration that changes URLs, strips structured data, moves content behind JavaScript or breaks the entity signals resets that confidence. Redirects protect link equity for search engines but they do not automatically carry citation confidence across, because a model may simply not re-crawl and re-verify for weeks or months.
Why is AI citation loss harder to spot than ranking loss?
Rankings are tracked daily and a drop is visible within days. AI citations are not tracked by most businesses at all, and the loss arrives gradually as each platform re-crawls on its own schedule. The typical pattern is that nothing looks wrong for six to ten weeks, organic traffic holds up because redirects are working, and then referral traffic from assistants quietly stops without anyone connecting it to the launch three months earlier.
When should we call you in a migration?
Before the URL structure is signed off, which usually means four to six weeks before launch. That is when the decisions that matter are still cheap to change. The next best moment is once staging is up but before content is locked. Calling after launch is still worth doing, because recovery is possible, but it is slower and more expensive than prevention and some of it depends on third-party sources re-crawling on their own schedule.
What does migration protection cost?
The pre-launch baseline and staging review are normally scoped as a Synaptic Audit at £2,950, which is credited in full against your first retainer month if you continue within 30 days. Launch-window monitoring and the 30 and 90 day recovery checks run inside a retainer, from £2,950 a month on Foundation. The exact shape depends on how close the launch date is and how many URLs are moving, so it is scoped rather than sold off a shelf.
Can you work from a staging site under NDA?
Yes, and it is the normal way this runs. We work from a staging URL with basic auth, under NDA, alongside whichever agency or in-house team is doing the build. We do not need to touch the production site to do the baseline or the staging review, and we do not need to be the ones deploying.
Will redirects not handle this?
Redirects handle the part search engines care about, which is passing link equity from an old URL to a new one. They do not handle the parts AI systems care about. A redirected page still has to be re-crawled, re-parsed and re-trusted, its structured data has to survive the move, its answer-first structure has to survive the CMS change, and the third-party sources citing the old URL have to be found and updated. A migration with perfect redirects can still lose citations.
How late is too late to call?
There is no point at which it becomes pointless, but the value curve is steep. Before URL sign-off is the best case. Before content lock is good. Launch week is reactive but useful, because catching a blocked crawler or a stripped schema in the first days avoids months of decay. Post-launch is recovery rather than protection, which is a longer and more expensive piece of work with a less certain outcome.
Do you replace the agency doing the migration?
No. We run alongside them, and often the agency is the one who brings us in. A development or design agency doing a replatform has a clear commercial interest in the launch not quietly costing the client their AI visibility, and no particular reason to have the specialist capability in-house. That arrangement can run white labelled under the agency’s brand if preferred.
Start Before The Date, Not After It
If you have a launch date, the useful next step is the baseline, because it is the only part that becomes impossible later.
- Book a conversation with the launch date and we will tell you what is still worth doing in the time available, including if the answer is “not much, wait and do the 30 day check instead”.
- Run the free AI Visibility Score on the current site first. It takes a minute and tells you whether there is anything worth protecting.
Related reading: how to measure AI search visibility for what a baseline should contain, why am I not in AI search results if citations have already gone, and GEO agency UK for how this fits the wider programme.