We analysed 1,200 visibility drops. Most had one cause
Across six engines and three months of data, most significant drops traced back to a single competitor page that an engine started citing.
Search engine optimization has always been about a list: ten links, and the goal is to be near the top. Answer engines produce something else entirely: a paragraph that names a few brands, explains why, and cites a handful of sources. There is no position eleven. If ChatGPT recommends three warehouse tools and yours is not one of them, the buyer will never see it.
That changes the objective. The question is no longer “where do we rank” but “are we in the answer, in what position, in what tone, and which page did the engine trust to say it”.
A page that ranks for a keyword may never be cited by an engine. Engines are looking for text that answers the exact prompt a user typed: a comparison, a best-for list, a direct yes or no with the reasoning. Pages that mention the keyword thirty times without answering anything tend to be skipped in favour of a Reddit thread that does.
In practice this means headings written as the questions people ask, a direct answer in the first sentence under each heading, and sources an engine can verify.
Ask the same question at 9am and 3pm and the list changes. That is not a bug in your tracking; it is how generation works. Any serious GEO measurement runs each prompt many times a day and reports how many runs the number is based on. A single screenshot from a colleague is an anecdote.
Most GEO work falls into three buckets. Content: pages that answer specific prompts and cover the comparisons you are losing. Earned: getting mentioned on the sources engines already cite, such as review sites and community threads. Technical: making sure AI crawlers can read your site at all, which is where a surprising number of brands fail with a 403 in robots.txt.
The order matters. Technical fixes are hours of work for multi-point gains. Content is days. Earned is weeks. Start at the cheap end.
Across six engines and three months of data, most significant drops traced back to a single competitor page that an engine started citing.
Every number in Askority now says how many runs it is built on, and every move is labelled before you react to it.
A checklist for finding out whether GPTBot, PerplexityBot and ClaudeBot can reach your pages, and what to change when they cannot.