When AI gets slower the longer you chat, first open a new conversation and restate the key background. In most cases the speed comes straight back, because the computing cost of a long conversation keeps piling up, and a new conversation resets it to zero. If the new conversation is still slow, check the three places below.
Step 1: Shorten this conversation
Close long conversations you no longer need, or save them as notes, and carry only the necessary background into a new one: the goal, key figures, and conclusions already settled. Pasting tens of thousands of characters of material at once and then following up forces the model to reprocess all of it every round, so slowness is normal. With a lot of material, have it processed in sections, or paste only the section relevant to the current question.
Step 2: Tell model slowness from network slowness
If the page keeps spinning and the prompt cannot even be sent, check the network and the product status first: switch networks, refresh the page, and try again at another time. If only a certain time of day is slow, it is probably peak-hour queuing, not your problem. If text comes out in small bursts but very slowly, it is usually the model's generation speed; switching to a lighter model or asking for a shorter answer often helps immediately.
Step 3: Do not let answers keep getting longer
Asking for everything to be "explained in detail" and "all angles covered" naturally produces longer answers and longer waits. Get a short answer with the key points first, and only dig into details once the direction is right. Turning on web search also lengthens the wait; for questions that are not urgent, try switching it off first.
If things are still slow after a new conversation, a network change, and shorter requests, try a different product or model once. That immediately shows whether the issue is the account, the product, or one particular model. Anyone who chats in long conversations every day should build a habit: one new conversation per task, save key conclusions as notes in good time, and do not pile everything into the same dialogue.
