Read the movement before you react to it.

When people ask why their ChatGPT citations changed, they usually assume something broke on their website. Far more often, OpenAI changed the model. An answer engine does not keep a static index of who to cite; each model version re-decides, at answer time, which sources to retrieve, how many to include, and how to weigh them. Change the model and you change all three at once — for everyone, on OpenAI's schedule, not yours.

The clearest evidence is the switch to GPT-5.5. On 23 May 2026, SISTRIX recorded a 47% citation shift within 48 hours of the model identifier changing, meaning almost half of all cited domains were different after the update than before. The average number of cited sources per response dropped from 30.9 to 28.4, and Reddit alone gained citations by about 59%. On a normal day in that tracking period, citation patterns moved only 1–2%. SISTRIX calls these events "ChatGPT Core Updates," by analogy to Google Core Updates — and is careful to note the data shows correlation, not proof that the model change caused the movement.

It is not a one-way street either. seoClarity, tracking five markets between February and May 2026, found ChatGPT's citation volume fell by over 90% at a March–April trough and then rebounded toward pre-March levels in May. What looked like a permanent decline turned out to be volatility: sharp moves in both directions. That is the single most important thing to internalize before you touch anything — a drop is often not a verdict on your content, and a recovery is often not a reward for a change you made.

The next model is already in motion. OpenAI previewed GPT-5.6 (the Sol/Terra/Luna family) on 26 June 2026 as a limited, government-coordinated release, with general availability promised in the following weeks. When that GA reaches the consumer ChatGPT surface, expect another reshuffle of who gets cited — a jump and a reordering, not a smooth trend. This guide is written so you can read that movement correctly when it lands.

Citation movement has more than one cause, and they call for opposite responses, so the first job is to name which one you are looking at. A platform change is a new model or a retrieval tweak from OpenAI; it moves everyone at once and is nothing you did. A category change is when a whole topic gets re-weighted — for example, ChatGPT leaning harder on community sources, which is why Reddit surged at the GPT-5.5 switch. A competitive change is when a rival earns more citations and displaces you on the merits. An owned-site change is the only one you caused: you edited a page, changed your robots rules, broke your structured data, or shifted your entity signals.

The reason this matters is that reacting to a platform change as if it were an owned-site problem is how brands waste a quarter. If OpenAI reshuffled the citations for a whole category, rewriting your page teaches you nothing, because the movement had no relationship to your content in the first place. Conversely, if a competitor genuinely out-evidenced you, blaming "the algorithm" lets a fixable gap sit open. The diagnostic steps below exist to sort these apart quickly, before you commit effort to the wrong one.

Two structural facts make ChatGPT especially prone to visible swings. First, it is the most parsimonious major engine, citing roughly 3.7 sources per answer against Gemini's 11.0 and Perplexity's 8.6 — with so few citation slots, small model changes produce large visible turnover. Second, ChatGPT only runs a live web search on about 34.5% of queries, so much of what it says never touches a citation at all. Both facts mean the citation surface you are watching is narrow and easily perturbed, which is exactly why a time-series view beats a snapshot.