The same question, asking ChatGPT today, and asking three months ago, may not be half the brand list. It's not your illusion, it's not the model that's gone stupid -- the reference to the generator engine. The source would have drifted, and it would have been large enough to change the list of candidates for a purchase decision. We spent a season tracking the same set of problems, trying to see how big the wave was, who it was, whether there were rules. The conclusion is that the wave is far more powerful than most brands think, and that it is better than a single ranking to reveal your true place in the eyes of AI.
What are we tracking?
The practice is simple. We've picked up 40 B2B buyers who really ask questions about the "best XX software" "XX tool" and the typical meaning of "XX for which team" and then ask each week the same words, either on ChatGPT, Perplexity or Gemini, and follow them for three months. Each time, it records which brands appear in the answers, which URLs are quoted and who is placed first.
Control change is the key to whether this investigation will work. We set up the same accounts, the same areas, and shut down personal memory and search history, so that the changes in the answers come from models and sources themselves, not from our own traces. Questioning words are completely static, because if one word is changed, the content will be different, then it will become two questions, not a piece of time.
Three months later, the answer changed.
Let's start with the most anti-intuitive point: Wave is not just about cold door problems. Even the "best project management software", the list of names quoted. In three months, nearly 40 percent of the population was replaced. You think you've been holding the position, but you've been relaxed.
- At the beginning of the month and at the end of the month, there was an average overlap rate of about 55%, in other words, one of every three quoted brands was replaced within a season.
- About a third of the problem was that the first brand was mentioned at least once in three months.
- Perplexity is the most dynamic because it is highly dependent on immediate search results; ChatGPT is stable vis-à-vis the brands that have been created, but it changes as well.
- Long-tailed, Leager-type problem list shuffles at about two to three times the number of hot issues.
- New content that is well structured, well-defined and defined can be squeezed into the reference list two or three weeks after it is online.
The reference to stability is more reflective of true status than a single ranking -- ranking can be pushed up by a short article, but stability is quoted, and representative models have repeatedly determined that you are a trusted source.— Tenten GEO Brand Radar tracking observation
Why does the same question give different answers?
Waves come from several layers of variation. The bottom level is a search: the model immediately searchs or searchs the index before answering, and the search engine itself is reordered every day, with new pages coming in, old pages falling out, and the selected content drawn. The first layer is the model's own update, and the supplier fine-tunes the weight and security strategy every few weeks, and the same hint may trigger a different selection preference. In addition to the randomity of the generation, even if the first two layers remain intact, there will be small swings in words and examples.
There's another factor that can be ignored: your competition is moving. When an opponent issues a well-structured, well-defined comparative article, it may take your place in two or three weeks. AI's answer is a constant recalculated balance, not a final list.
There are rules behind the wave.
There's a predictable part of the mess. High-profile brands, backed by third parties, are clearly smaller; information books, brands supported only by their own page are the easiest to fall out of the shuffle. The clearer the intentions and the more common answers, the more stable the questions are; the more vague, dominant and multi-optional the questions are, the longer they are. This means that you can push yourself to the low-wave, stable, quoted end with content structures and external expertise.
How does a brand respond to this instability?
The least effective response is to intercept AI every day and to panic about an article when you see yourself falling off. A single-day list would've been shaking, and a single-point decision would've only cost you your life. What really needs to be done is to lengthen the period of tracking, to determine whether it moves to a steady or continuous loss, using such indicators as the overlap rate and the amount quoted.
To improve stability, the direction is clear: to write the core theme into clear, numerical and contextual paragraphs that allow models to be extracted cleanly; to maintain consistent brands on multiple credible sources Physical signal; and build up third-party references and corroboration to allow AI multiple reasons to make you a credible source. And that's exactly what Tenten's Brand Radar is doing -- continuously monitoring your reference tracks in various home-generated engines, filtering out one-day messages into a decision-making trend.
The answer to AI will not stabilize you, and you will only make yourself a source that can't be replaced. If you want to know how stable you're being quoted and where the gaps are, you can schedule a 30-minute GEO diagnosis, and we'll run a track on your own brand and talk directly to the numbers.
