Why coverage differs
Three reasons. Training data: each model learned from a different corpus at a different time, so your brand may be well represented in one and thin in another. Retrieval: when assistants search live, they favor different sources and parse pages differently. And reasoning style: given the same evidence, models weigh community sentiment, editorial reviews, and structured data differently when building a shortlist. The sourcing mechanics are covered in where ChatGPT gets its information.
What to do about it
Measure all four, not one. A brand that only tracks ChatGPT routinely discovers it is invisible in Gemini or misdescribed in Perplexity. The fixes overlap heavily, direct answers, schema, consistent facts, trusted citations, so closing a gap in one usually lifts the others, but only measurement tells you where the gaps are. The free checks ChatGPT, Gemini, Claude, and Perplexity together, and per-assistant tracking guides live in our ChatGPT, Gemini, and Perplexity monitoring pages.
Running a cross-assistant audit this week
The audit takes one hour. Pick ten questions your buyers actually ask, in their words, including at least three that describe a situation rather than a product. Run all ten through ChatGPT, Gemini, Claude, and Perplexity, and build a simple grid: question by assistant, with the brands named in each cell. Your gaps will cluster: usually one assistant where you barely exist, and a set of situational questions where nobody has claimed the answer yet. Those unclaimed moments are your fastest wins, because you are competing with absence rather than an incumbent.
Re-run the same grid monthly and the trend tells you more than any tool score: which fixes transferred across assistants, which assistant lags, and whether competitors are gaining in the moments you care about. When the grid gets too big to run by hand, that is the handoff point to software, covered in do I need software for GEO and AEO.
Reading the gap: what each pattern means
Once your grid exists, the patterns diagnose themselves. Strong in ChatGPT, weak in Perplexity usually means your presence relies on trained knowledge while your live, citeable pages are thin, Perplexity leans hardest on retrieval and shows its sources. Strong in Perplexity, weak elsewhere is the reverse: good pages, not yet enough accumulated reputation for models to know you without looking. Weak only in Gemini often traces to your Google surface, business profiles, structured data, and the pages Google indexes best. And uniformly weak everywhere is not four problems, it is one: the shared input layer, which is the cheapest possible diagnosis because one fix program lifts all four. The worst response to any pattern is per-assistant content hacks; the input layer explains eighty percent of gaps, and where ChatGPT gets its information explains why.
Prioritizing when you cannot fix everything
Weight the grid by where your buyers actually are. A developer-tools brand can often deprioritize an assistant its audience barely uses; a consumer brand cannot ignore any of the big four. Multiply roughly: assistant usage in your audience, times the purchase intent of the moments where you are weak, times the size of the gap. Fix the biggest product first, which is frequently not the biggest brand-name assistant but the one your highest-intent moments run through. Re-run the grid monthly and expect convergence: as the shared layer strengthens, the assistants agree about you more and more, and the remaining differences become genuinely surface-specific, X for Grok, communities for others, at which point per-surface investment finally makes sense. The tracking pages for ChatGPT and Gemini cover the per-assistant loops.
A quarterly convergence review worth doing
Beyond the monthly grid, run one deeper review each quarter: convergence analysis. Compute, roughly, how often the four assistants agree about your moments, all naming you, none naming you, or split. Rising agreement in your favor is the strongest signal the shared input layer is winning, because four independent readers reaching the same conclusion means the evidence, not any single model's quirk, is carrying you. Persistent splits are your remaining surface-specific work, and the quarterly view catches slow drifts the monthly grid hides, an assistant quietly demoting your category's comparison sources, a competitor's community push landing on one surface first. Twenty minutes with a spreadsheet, once a quarter, and cross-assistant strategy stops being anecdotes and becomes a trend you can manage, in both directions.
