Perception Drift Tracking

In January, ChatGPT calls your brand the 'leading enterprise platform'. In April, the same prompt returns 'a popular tool for small teams'. Nothing about your product changed. What changed is the source mix the model retrieved from, a few new blog posts, a couple of forum threads, and one well-ranked review reframed the story.

Drift rarely happens in one big jump. It accumulates. Each individual change looks insignificant; the compounding effect is brand-defining.

Brand teams check AI responses ad-hoc, once a quarter, once before a board meeting. By the time someone notices the description is off, the drift has already shaped dozens of buyer conversations.

Traditional brand-tracking studies catch this for the human market with quarterly surveys. AI moves faster. Perception in models can shift in days, not quarters.

We run a fixed prompt set against all 8 AI models every week. Each response is scored across 40+ perception metrics, positioning match, category framing, sentiment, target-customer language, competitor proximity, sources cited. Scores are versioned so week-over-week deltas are explicit.

When a metric crosses a threshold, the dashboard surfaces a drift alert with the exact wording change, the model that changed, and the new sources it started citing.

Reversing drift is a source operation, not a copywriting one. Identify the URLs the model started citing for the new framing. Most of the time, a small number of pages (3–7) carry disproportionate weight. Updating, correcting, or counter-publishing against those pages is what changes the next week's response.

Pair source work with consistent positioning across your owned channels. The model retrieves; consistency is what survives retrieval.