Why every new ad task should start in a new chat

AI carries the last meeting into the room with it

Monday morning, you dug into last week’s ROAS drop. In the same chat, you cleaned up copy for a client report on Tuesday, then asked for a budget draft for a new campaign on Wednesday. But the AI keeps answering as if it’s still Monday — pinned to GA4 as the source, a week-over-week comparison, and the campaign exclusions you set that day.

The problem here isn’t that the AI suddenly got dumber. It’s that the assumptions you stacked up earlier in the conversation are still in play on the next task. That’s especially risky in ad operations. Mix up date ranges, ROAS sources, campaign objectives, or approval status even slightly, and you get answers that sound plausible but fall apart the moment you try to act on them.

Chat context is more like working memory

Most AI chats feed everything in the current conversation — your messages, the AI’s answers, files, and tool outputs — into the next answer. OpenAI notes that as a conversation gets longer or runs many turns, you have to account for the context window and token limits. Anthropic likewise describes the context window as the model’s working memory: what it can pull from when it generates a response.

The key point: more context isn’t always better. When you’re going deep on a single advertiser, a long thread helps. But the moment you switch to a new advertiser, a new time frame, or a new decision, that old thread turns into noise. Sometimes keeping only what’s relevant to the task at hand beats giving the AI more to chew on.

A new chat isn’t a totally blank slate, either. Depending on the product, stored memory, project guidelines, workspace guidelines, and Gem guidelines can still carry into new conversations. So in practice, this split is safer: keep your team’s standing rules in a Gem or in workspace guidelines, and spell out this task’s dates, objective, and decision criteria in the first message of the new chat.

Split by the decision, not the question

A simple rule like “same chat while the questions keep flowing, new chat when the question changes” doesn’t go far enough. In ad operations, it’s better to split by the decision than by the question. Keep narrowing down the root cause on the same campaign? Same chat. But when you move from analysis to drafting an action plan, or from report copy to reviewing a budget change, a new chat is the safer call.

The contexts that are most dangerous to mix in ad operations

Context contamination doesn’t always blow up into a major incident. But in ad operations, where real spend and client relationships are on the line, a few contexts have to stay separate. It’s especially easy to miss when the AI takes a numeric threshold from an earlier conversation and applies it straight to the current task.

How Neuro draws the line

In Neuro, opening a new chat means kicking off a new unit of work. It doesn’t throw away your team’s operating standards. Your workspace guidelines, connected ad platforms, and selected Gems stay in place as standing rules — only this task’s goals and decision criteria get declared fresh.

1. Pick the objective first. Pin it to one of: diagnosis, reporting, action-plan review, approval request, or a how-it-works explanation.
2. When the objective changes, open a new chat — especially when you move from analysis to execution or client comms.
3. Choose the Gem that fits the task. Lock in its role: a weekly report Gem, a budget review Gem, a creative review Gem.
4. In your first message, spell out the time frame, data source, comparison basis, exclusions, and the outputs you want.
5. Don’t ask for conclusions right away. Have the AI summarize this chat’s assumptions first.
6. When the analysis is done, carry only the summary you need into the next chat. Don’t copy-paste the whole thread.

When it’s fine to keep going in the same chat

The point isn’t to spin up as many new chats as possible. When you need to keep refining the tables, hypotheses, or calc rules the AI just produced — all within the same decision — the same chat is better. Cutting context there just runs up the cost of re-explaining everything.

• Digging deeper into the root cause of a performance drop on the same campaign
• Tightening or polishing report copy the AI wrote under the same criteria
• Swapping only the segments while the time frame and data source stay the same
• Asking for extra supporting tables for a hypothesis it just put forward
• Shoring up the rationale on an approval card, or laying out the reasons for a rejection

The rule is simple. If the last answer needs to be raw material for the next one, stay in the same chat. If the last answer could bias the next one, open a new chat.

First-line examples for a new chat

The first line of a new chat doesn’t need to be long. It just needs to draw a clear boundary around the task. You can paste the lines below straight into Neuro and only swap in your own dates, advertiser, and campaign names.

What belongs in your team rules

A good AI operations team isn’t the one writing long prompts every time. Standing standards live in Gems and workspace guidelines; the task context that changes every time goes in the first line of a new chat. Split it this way and the AI gets confused less often — and when teammates read the chat logs, it’s far easier to trace which decision was made under which criteria.

• One chat, one decision.
• Put the advertiser, time frame, and objective in the title of every new chat.
• Keep standing team standards in Gems or workspace guidelines, not buried in the chat itself.
• On the first reply, have the AI summarize this chat’s assumptions before it draws any conclusions.
• When you’re unsure whether to keep going or start fresh, check whether the objective, time frame, data source, permission level, or audience has changed.
• When you move to the next task, carry only the final summary, the confirmed criteria, and the open questions — not the whole conversation.

The quality of AI ad operations isn’t set by the model alone. It comes down first to operating habits — what you hold onto and what you start over. A new chat isn’t just good housekeeping; it’s a safety mechanism for your decisions.