Key takeaways
- Treat a platform switch like a re-baseline, not a lift-and-shift. Citation behavior changes on its own (ChatGPT's average citations per response dropped roughly 27% after the GPT-5.3 rollout in March 2026), so you can't assume last month's Searchable numbers and this month's new-tool numbers are measuring the same thing.
- Run the old and new tool in parallel for 2-4 weeks before you cut over reporting. A single week of divergence is normal; a sustained gap tells you something real.
- Rebuild the prompt set, don't copy-paste it. Export what you had, but re-weight it against current buyer language, fanout patterns, and the new tool's prompt limits.
- Separate the "why" conversation from the "what changed" conversation in your first client update. Clients panic over a visibility score moving 8 points; they calm down when you show them it's a measurement-layer change, not a performance drop.
- Document everything (prompt list, cadence, source weighting, reconciliation notes) in a shared onboarding file. The next platform switch, whenever it happens, will take a fraction of the time.
Agencies switch AI visibility vendors for the same reasons they switch any martech tool: pricing changed, a feature they needed shipped somewhere else, seat limits started punishing growth, or a new account lead inherited a client and wanted a tool they trust. Worth a quick correction before we go further: Searchable itself hasn't shut down. It's an active platform that recently shipped a "Prompt Universe" feature and runs an agency partner program. If your agency is "leaving Searchable," it's almost certainly a fit decision, not a forced migration, and that distinction matters because it changes how you frame the move to the client. You're not explaining a vendor failure. You're explaining an upgrade.
That framing matters more than it sounds like it should. Clients don't care which dashboard you use. They care whether the numbers they've gotten used to are still trustworthy after the switch. Get the onboarding wrong and you'll spend the next quarter answering "why did our score drop 15 points" questions that have nothing to do with actual visibility.
Step 1: Audit what you're actually leaving behind
Before touching the new tool, pull a full export from the old one. You need:
- The exact prompt list, including any that were added or retired mid-engagement
- Historical citation/mention data for at least the trailing 90 days
- Any custom tagging (competitor sets, geography, persona labels)
- Screenshot or PDF copies of past client reports, since raw exports rarely preserve the formatting a client has gotten used to
This sounds obvious, but agencies skip it constantly because the old tool is already half-forgotten by the time the new contract is signed. Don't let that happen. The historical data is the only thing that lets you answer "is this better or worse than before" six months from now.
The Wellows onboarding framework boils this down to three questions you should be able to answer before you log into the new platform at all: how does AI currently perceive the brand, where is the brand visible or invisible, and what will actually move the needle. If you can't answer those from the old tool's export, your migration data is incomplete.
Step 2: Rebuild the prompt set, don't migrate it as-is
The instinct is to copy the old prompt list straight into the new tool. Resist it. Two things have probably changed since that prompt set was built: the client's product or positioning, and the way AI engines actually search.
On the second point, the data is pretty stark. Promptwatch's tracking of ChatGPT query fanouts shows the average number of searches triggered per response falling from roughly 2.15 in early December to exactly 1.0 by April, while the average internal query length dropped by more than half, from around 117 characters down to roughly 53. ChatGPT isn't asking itself long conversational sub-questions anymore. It's firing off short, keyword-style retrievals. A prompt set designed around 2025-era conversational phrasing is already stale.
Practical move: pull the old prompt list, cross-reference it against current SEO keyword data and real support/sales language, and cut anything that hasn't surfaced a meaningful signal in 90 days. Profound's prompt design methodology (one of the more documented approaches in this space) follows a similar logic: start from existing keyword inputs, convert them to conversational question format, then layer in real customer language before finalizing. Whatever tool you land on, do this step manually rather than trusting an auto-import.
Also check the new platform's prompt cap against what the client actually needs. Entry tiers vary a lot. Searchable's own Pro plan caps at 100 prompts and one project, which is exactly the kind of ceiling that pushes multi-brand agencies into a pricier tier. Know your new tool's limits before you build the prompt list, not after.
Step 3: Run both tools in parallel for a reconciliation window
This is the step agencies skip most often, usually because of budget pressure (why pay for two tools at once?) or impatience (the client wants the new dashboard live). Skip it anyway and you'll regret it the first time the client asks why visibility "dropped" the week you switched.
Run the old and new platform side by side for two to four weeks. You're not trying to get identical numbers. Different tools sample at different cadences, weight sources differently, and sometimes track different engines by default. You're trying to establish a reconciled baseline: a range you can defend when a client asks "so are we actually visible or not."
This matters even more right now because citation behavior itself is unstable at the platform level. Around the GPT-5.3 rollout on March 4, 2026, average citations per ChatGPT response dropped from about 6.4 the week before to roughly 4.7-4.9 by late March, a platform-wide change that hit every GPT-5.3-era model simultaneously, not a quirk of one tool's measurement. If your reconciliation window happens to straddle a rollout like that, you need to know it's a platform event and not a tool-switch artifact, or you'll misattribute a real drop to the migration.
Promptwatch's own data on citation volume backs this up from a different angle: ChatGPT cites around five sources per web-search response, versus roughly ten for both Google AI Overviews and Perplexity, and Microsoft Copilot has swung from under two to nearly seventeen sources per response within a matter of weeks. If you're reconciling numbers across tools and across engines, you need a baseline that accounts for the fact that each engine's citation inventory behaves completely differently. A one-week snapshot on Copilot tells you almost nothing; a one-week snapshot on Perplexity is fairly stable.
Step 4: Set the reporting cadence before the first dashboard goes live
Clients remember whatever cadence they got used to with the old vendor. If Searchable (or whichever platform you're leaving) delivered a monthly PDF, don't suddenly switch to a raw weekly Slack dump and expect them to adjust quietly. Decide the cadence up front and communicate it as part of the migration, not as a surprise.
A tiered structure works well for most retainers:
- Daily or near-real-time: internal dashboard access for the account team, not the client
- Weekly: a short digest flagging anything that moved outside the normal range, with hallucinated citations and one-off noise filtered out before it reaches the client
- Monthly: the client-facing report, tying citation share and mention rate back to branded search lift or verified leads where you can
- Quarterly: a strategic review that revisits the prompt set and resets the baseline if the client's positioning or market has shifted
This cadence structure also gives you a natural place to handle the single biggest trust risk in any migration: reporting noise that looks like a real signal. A one-day dip in visibility is usually retraining, a model version swap, or random sampling variance, not an actual loss of ground. Build the rolling average into your weekly digest so the account team isn't making panicked calls over normal volatility.
Step 5: Brief the client on what changes and what doesn't
The client-facing conversation should be short and specific. Three things to cover:
- The new tool tracks engines slightly differently (name the engines added or dropped)
- There's a 2-4 week reconciliation window where numbers may look different from last month's report, and that's expected, not a performance signal
- The historical trend line will be preserved (show the old data stitched to the new baseline), so they're not starting from zero
Clients who understand that volatility is normal are much easier to manage long term. One of the more sobering stats here, from AirOps's State of AI Search research, is that only 30% of brands stay visible from one AI answer to the next, and only 20% remain visible across five consecutive runs. If a client doesn't know that going in, every dip looks like a crisis. If they do know it, a dip is just Tuesday.
Comparing where agencies land after leaving a mid-market tool
If you're moving off Searchable or a similar mid-market tracker, here's roughly how the field breaks down by workflow fit rather than feature checklist:
| Platform | Entry price (approx.) | Engines tracked | Best fit |
|---|---|---|---|
| Otterly.AI | $29/mo | 4 engines + add-ons | Fast client onboarding, non-specialist dashboards |
| Peec AI | €85/mo (brand), $245/mo (agency) | Up to 6 models | Teams that want smart prompt suggestions |
| Gracker AI | $99/mo | Up to 10 engines | Agencies wanting prompt mining and page-level audits |
| Profound | ~$399+/mo per client workspace | 3-9 engines depending on tier | Enterprise agencies needing SSO/SOC 2 |
| Promptwatch | $95-$579/mo (agency tiers from $199/mo) | ChatGPT, Gemini, Claude, Perplexity, Grok, DeepSeek, Mistral, Copilot, AI Overviews, AI Mode, and more | Agencies that want monitoring plus automated content execution, not just a dashboard |
Most of the field, including Searchable itself, stops at monitoring: prompts in, citations out, a dashboard to screenshot for the client. That's fine for a quick audit but it leaves the "now what do we actually fix" question to the account team's judgment.
Promptwatch is built differently for agencies that want to close that loop. Beyond multi-engine prompt tracking, it logs AI crawler activity (ChatGPTBot, ClaudeBot, PerplexityBot and 400+ others) so you can see which pages AI systems are actually reading, and it includes Content Agents that draft and publish GEO-optimized pages directly to a client's CMS based on the gaps the tracking surfaces. For an agency managing ten or twenty client accounts, that's the difference between handing a client a chart and handing them a prioritized to-do list that your team already started executing.
It's used by 1,840+ brands and agencies, including agency networks like Monks and WPP, and sits at the lower end of the cost-per-response spectrum once you account for how many engines and markets are included by default. Not every client needs that full stack on day one, but if the migration conversation is happening anyway, it's worth weighing against a straight prompt-tracker swap.
Common onboarding mistakes to avoid
A few patterns show up again and again when agencies migrate clients between AI visibility tools:
- Hallucinated citations treated as wins. If the new tool flags a citation it can't verify, pull it from the metric before it reaches the client. Don't celebrate a number you can't defend.
- Brand-name collisions. A client called something generic can pick up mentions meant for an unrelated company. Run a baseline week specifically looking for this before you trust any new tool's numbers.
- Single-tool dependency after the reconciliation window closes. For high-value accounts, keep a lightweight secondary check running even after you've fully cut over. One Reddit thread from a practitioner summed up the frustration well: manual prompt-checking doesn't scale past two clients, and tool-reported numbers are inconsistent enough that agencies need a reconciliation habit, not a one-time event.
- Skipping the quarterly prompt review. Categories evolve. A prompt set that reflected real demand six months ago drifts out of relevance, especially as fanout behavior keeps shrinking and sharpening across engines.
Wrapping up the migration
The actual software switch is the easy part. Most AI visibility platforms can get a client's domain added and a basic prompt set running within a day. The work that actually protects the retainer is everything around that: the reconciliation window, the honest client briefing, the prompt set rebuild, and the documentation that makes the next switch faster than this one.
If your agency is building this into a repeatable service line rather than a one-off migration, it's worth browsing the broader GEO software directory at bestgeosoftware.com to see how the field is segmenting by price and feature set before you commit a client roster to any single vendor long term.