Most GEO plans react to what AI already says about you. A frame-first plan gets ahead of the next training cycle instead.
Here’s the problem with the way most teams run GEO right now. They check what ChatGPT or Perplexity says about their brand, notice something’s off, and scramble to fix it. That’s not a strategy. That’s damage control on a model that already made up its mind months ago.
By the time you notice you’re missing from an AI Overview or getting recommended third instead of first, the model has already learned a story about your category from sources it crawled a while back. You’re not fixing today’s answer. You’re trying to catch a train that already left the station, and the next one won’t arrive until the next training or re-indexing cycle.
A 90-day plan that actually works doesn’t start with “what does AI say about us.” It starts with “whose frame is the model running on, and how do we own that frame before it hardens again.”
Why 90 Days? The Model Training-to-Deployment Window
AI models don’t update their worldview daily. Foundation models retrain on cycles measured in months. Retrieval layers like AI Overviews, Perplexity, and Copilot refresh their source pools faster, but the narrative patterns they lean on (which sites they trust, which frames they repeat, which brands get cast as “the innovative one” versus “the legacy player”) tend to stick once they’re established.
Ninety days is roughly the window where two things overlap: it’s long enough to actually shift the sources and content that models draw from, and it’s short enough to land before the next major retraining or re-crawl cycle locks a new consensus in place. Wait longer and you’re optimizing for a narrative that’s already calcified. Move faster and you haven’t given the new sources time to get crawled, cited, and absorbed.
Think of it less like an SEO sprint and more like a narrative campaign with a deadline. You’re not chasing a ranking. You’re trying to be the frame the model adopts next time it looks.
Narrative Audit (Days 1–20): Map the Frame AI Already Learned
Start by asking the AI directly, across models. Try “best project management tools for remote teams” in ChatGPT, Perplexity, and Gemini, and read not just who gets named, but how each brand gets described. One tool might get framed as “powerful but complex,” another as “simple but limited.” That framing is the story the model has already learned, and it didn’t come from nowhere.
Your job in these first three weeks is to reverse-engineer that frame. Where did “powerful but complex” come from? Probably a mix of review sites, comparison posts, Reddit threads, and old category explainers that all repeated the same angle until it became consensus. Map it out: what’s the dominant narrative for your category, who’s cast as the hero of that narrative, and where does your brand actually sit in it. Often you’ll find the model has flattened your brand into a caricature, a version of you built from outdated or thin sources instead of who you actually are today.
Prompt Universe Mapping (Days 21–40): Find the Questions Shaping Recommendations
Buyers don’t search “best CRM.” They ask “what CRM should a 15-person agency use that won’t require a consultant to set up.” AI search runs on prompts like that, not keywords, and the prompt itself shapes which sources the model pulls from and which frame gets triggered.
Spend these three weeks building out the real prompt universe for your category: the comparison prompts, the “is X worth it” prompts, the “alternatives to X” prompts, the ones specific to your buyer’s job title or company size. For each cluster, check who gets recommended and why. You’ll likely find your brand shows up fine on generic prompts but disappears or gets misrepresented on the specific, high-intent ones, which are exactly the prompts closest to a buying decision.
Source Intelligence (Days 41–60): Identify Which Sources Drive the Narrative
Now trace the citations. When an AI answer names a competitor as the top recommendation, follow the trail: what pages is it pulling from, what review sites, what comparison content, what Reddit or forum threads keep getting surfaced. Some sources carry outsized weight because they’re crawled often, cited across multiple models, or treated as neutral third parties even when they’re not.
This is the step most GEO plans skip entirely, because a dashboard telling you “you’re mentioned 40% less than Competitor X” doesn’t tell you which three sources are actually responsible for that gap. Source intelligence does. Once you know which pages and publishers are shaping the frame, you know exactly where to focus.
Frame-Ownership Plan (Days 61–90): Ship Content That Rewrites the Consensus
With the audit, the prompt map, and the source list in hand, the last 30 days are about shipping, not planning. That means getting your framing into the sources the model already trusts (contributing to the comparison posts, correcting the outdated claims, showing up in the forums), and publishing your own content that directly answers the high-intent prompts you mapped in phase two, using the language your buyers actually use.
This isn’t about publishing more content. It’s about publishing the specific pieces that rewrite the two or three frames that are actually costing you recommendations.
Proof: Measuring Narrative Share, Not Just Presence
At the end of 90 days, don’t just check if you’re mentioned more. Check whose frame the model is repeating. If AI still describes your category the same way, with the same hero and the same caricature of you, presence didn’t matter. Narrative share, whose story the model actually adopts when it explains a recommendation, is the number that tells you if the work landed.
Beyond Day 90: Sustaining Frame Ownership
Ninety days gets you a new frame in motion. It doesn’t lock it in forever. Models retrain, competitors publish, sources get re-crawled. Treat this as the first cycle of an ongoing practice, not a one-time fix, and plan the next narrative audit before the current frame starts to drift.
If you want help running this without guessing at which sources actually matter, that’s what Mavel is built for. Come see what frame the models have already built around your category, and what it’ll take to own the next one.
