Ask two AI engines the same question about gyms in your city and you can get two different answers. One names you. The other has never heard of you. Owners find this maddening until they see the reason, and the reason is simple: the two engines are answering in different ways.
Understanding the split tells you exactly where to invest. This is the plain language version.
Mode one: grounded answers
Some engines search the live web before they reply. Perplexity works this way by default, and ChatGPT does it when it decides a question needs fresh information, which local questions usually do.
The engine runs a search, reads what it finds, and builds an answer from those sources, often with citations. The industry calls this grounding. Think of it as an assistant doing quick homework before speaking.
For your gym, grounded mode means the answer reflects what is readable about you right now. Your website, your Google Business Profile, the directories that list you, this week’s timetable. If a machine can fetch it and parse it, it can shape today’s answer.
Mode two: answers from memory
The same engines can also answer without searching at all, purely from what they absorbed during training: an enormous snapshot of text gathered over years. Ask a general question, or use an engine without live search, and you get this mode.
Memory mode has a built in lag and a built in bias. The model knows what was written about widely, repeatedly, and consistently in the past. A gym that has been reviewed, listed, mentioned in the local press, and talked about for years has a real footprint in that memory. A gym that opened recently, or kept to itself, barely exists there, however good its website is today.
If that bias sounds familiar, it should. It is one more reason assistants reach for the national brand, a pattern we unpacked in why AI recommends the chain, not your independent gym. The chain got written about for a decade. That coverage is now baked in.
What each mode rewards
Here is the practical difference, and it is worth writing on the whiteboard.
Grounded engines reward the state of your web presence today. Crawlable pages, structured data, a timetable in real text, hours that agree everywhere, an llms.txt file, current prices. Fix these and grounded answers can start including you within weeks, because the engine reads the improvement the next time it looks.
Memory engines reward being written about, widely and over time. Directory entries, local press, fitness media, neighbourhood guides, reviews with substance. You cannot inject anything into a model’s memory directly. You can only grow the body of text about your gym so that future training absorbs you, and so that grounded searches find third parties confirming you in the meantime.
One mode you influence in weeks. The other you influence in seasons. Both are worth it, because the same mention that feeds future memory also serves as a source for today’s grounded search. The work overlaps almost perfectly.
Where to invest, in order
Start with the grounded engines, because that is where effort pays fastest and where your own website is fully under your control. Make yourself readable: structure, consistency, currency. This is the technical checklist we covered in GEO for fitness businesses, and every item on it strengthens your grounded answers.
Then begin the slower channel without waiting for the first one to finish. Every month you delay mention building is a month of memory you never get back. Listings, local stories, genuine reviews, the occasional piece of fitness media. This is the compounding work behind reputation and signals, and the gyms that start it early become the ones the next generation of models simply knows.
And pay attention to which engines your own members actually use. Ask them at the front desk. A membership full of Perplexity users means grounded mode decides your new joins, so your website state matters most. A crowd that asks ChatGPT general questions leans more on memory, so your mention footprint carries more weight. Neither answer excuses you from the other channel; it only sets the priority.
What this means if your gym is new
Owners of young gyms often read all this and feel the deck is stacked, because memory mode belongs to businesses with a past. Flip that thought. A new gym has zero legacy errors in the record. No decade of outdated timetables, no old address haunting the directories, no closed-location reviews confusing the picture.
That makes the playbook unusually clean. Build the website right from day one, structured and current, and grounded engines can treat you as seriously as anyone in your postcode. Then let every month of operation add mentions: the opening covered by the local paper, the first wave of genuine reviews, the neighbourhood guide that includes you. You are writing the memory the next generation of models will hold, and you are writing it without competition from your own past mistakes.
Established gyms should run the same logic in reverse: before building anything new, clean up what the web already believes about you, because contradictions cost trust in both modes.
See which mode is failing you
The symptoms are diagnosable. Invisible everywhere usually means a readability problem on your own site. Visible in grounded engines and unknown in memory mode is the classic profile of a young, quiet independent: your site is fine and the wider web has little to say about you yet.
The instant way to test the first half is the free AI visibility check, which reads your website the way a crawler does and scores the signals grounded engines depend on. For the whole picture, request a full audit: we put real member questions to the main engines, in both modes, and show you exactly where your name appears, where it should appear, and which channel needs the work.
Two modes, two clocks. Win the fast one this quarter, feed the slow one every month, and the answers start agreeing about you.