Switching GEO tools without losing your history

Ask a team why they stay on a GEO tool they have outgrown and the answer is rarely the feature set. It is the graph. Eight months of visibility trend, the citation history behind a client report, the before-and-after of every content push: walk away from the tool and you walk away from the evidence. Vendors know this, which is why exports are usually an afterthought and imports usually do not exist.

We built the import. If you are moving a brand from Peec to llemmy, this is the playbook, including the two things most migration guides skip: the export cap you will hit on a long history, and why the numbers in your new tool will not match the old ones digit for digit.

What actually needs to move

A GEO workspace has three kinds of state, and they matter in different ways.

• The raw answers. Every AI response your prompts collected, with its date, model, cited sources and text. This is the irreplaceable part: you cannot re-ask ChatGPT what it said in February. Everything else can be rebuilt from it.
• The prompt set. The questions you track, plus the taxonomy around them: topics, tags, intent. Rebuilding this by hand is tedious and error-prone.
• The competitor roster. Which brands you measure share of voice against. Small, but it defines every comparative number.

Peec exports all three from its own UI as JSON, and llemmy imports each one: the All Chats export carries the raw answers, the Prompts export carries the taxonomy, and the Top Brands export seeds your tracked competitors.

The migration, step by step

1. Set up your brand profile first. In llemmy, enter your primary brand before importing. Branded prompts (the ones that name you) are classified at import time and kept out of core visibility, the same rule llemmy applies to its own data. Import first and you would classify against an empty brand.
2. Export from Peec. On the Overview page, pick a date range, then export the All Chats table as JSON. Do the same for Prompts and Top Brands if you want them.
3. Mind the cap. Peec writes at most 10,000 chats per export file, silently. A brand tracked daily across a few models hits that in well under two months, so an "all time" export of a long history is actually just the most recent slice. Export month-by-month ranges instead. Overlap between slices is fine.
4. Import. In llemmy, go to Settings, then Import, select every chats file at once (plus the optional prompts and brands files) and start the import. It streams in batches with a progress bar and reports exactly what landed: responses, prompts, engines, competitors.
5. Enable the prompts you want tracked. Imported prompts arrive paused so a migration cannot silently kick off hundreds of live runs. Turn on the ones llemmy should keep asking, and your history and your ongoing tracking meet on one timeline.

Why re-imports cannot double-count

Every imported response keeps a stable identity derived from the source record. Import the same file twice, or two files whose date ranges overlap, and the duplicates are recognized and skipped, with the count reported back to you. This is what makes the slicing workaround for the export cap safe: you do not need to cut ranges precisely, you just need to cover the period. If an import fails halfway on a flaky connection, retry it; what already landed is skipped.

The honest part: your numbers will move

Import your history and llemmy's visibility figure for last month will be close to what Peec showed, but it will not be the same number. This surprises people, so it is worth being precise about why.

• Different counting rules. Tools differ on what counts as a mention, how a brand name is matched, and how models are weighted into a headline rate. Neither answer is wrong; they are different definitions applied to the same text.
• Different denominators. llemmy excludes branded prompts from core visibility because a prompt that names you inflates the rate. If part of your imported prompt set is branded, the core number is computed on the unbranded remainder.
• Intervals, not integers. llemmy reports headline rates with a sample size and a 95% confidence interval. A month where one tool says 45% and the other says 41% with an interval of plus or minus five is not a discrepancy. It is the same underlying reality, reported with its uncertainty made visible.

The practical advice: after importing, treat the trend shape as the continuity, not the absolute level. What you keep is the ability to say "here is our trajectory over eight months, and here is what moved it," with the raw answers behind every point.

What carries over, and what does not

Carries over: every answer's text, date, model, cited sources, your prompt taxonomy, and your competitor roster. Recomputed by llemmy on top of it: visibility, share of voice, position, sentiment and citation analytics, so your history is measured with the same rules as everything llemmy collects for you going forward. Not carried over: Peec's own derived scores, since pinning another tool's computed numbers onto a timeline measured differently would make the graph lie.

One graph, one methodology, no blank start. That is the whole point.