When to start a new chat for ad work
Keep the conversation when you are reviewing the same campaign. When the account or objective changes, decide which assumptions should carry over.
AI carries the last meeting into the room with it
Suppose Monday’s chat reviewed last week’s results in GA4. On Wednesday, you use it to plan a new campaign budget. The new task needs platform-reported results, but the answer may still follow Monday’s dates and exclusions.
Split conversations by the work and its assumptions, not by the calendar. Moving from analysis to a proposed change for the same campaign benefits from keeping the evidence. A different advertiser or objective is a reason to establish a fresh context.
Chat context is more like working memory
Dates, metric sources, and excluded campaigns can become assumptions for the next answer. Before changing tasks, ask AI to summarize the conditions it is using, then say which ones still apply.
A new chat can still use saved Gems, workspace instructions, or memory. It is not an access-control boundary or a way to delete stored information. For another advertiser, check the selected workspace and connected account too.
Continue or start fresh?
| Situation | Conversation | Check first |
|---|---|---|
| Reviewing a budget proposal after analyzing the same campaign | Continue the chat. | Recheck current values and the exact execution target. |
| A different advertiser, account, or objective | Check the workspace and start a new chat. | Leave the previous account’s numbers and assumptions behind. |
| Same campaign, different reporting basis | State the new assumptions; split if they keep getting mixed. | Specify the source, dates, and exclusions. |
Refresh the facts before executing
Before executing a change, refresh the current account state even if you stay in the same chat. The budget or status may have changed since the analysis. Request execution only when the target and values still match what you reviewed.
First-line examples for a new chat
Sources
- OpenAI API Docs: Conversation state
Official OpenAI docs on managing the context window and token limits as conversations grow, noting that both input and output tokens count toward context usage.
- Anthropic Claude Docs: Context windows
Official Anthropic docs describing the context window as working memory the model can reference when generating responses, and explaining how conversation turns accumulate.
- OpenAI Help Center: Memory FAQ
Official OpenAI help article on how stored memory and references to chat history can carry into new conversations, and how users can manage their memory settings.
- OpenAI API Docs: Prompting
Docs on prompt management principles: put recurring instructions like role and tone in the system message, and put task-specific details and examples in user messages.
Related posts
How to use Gems in Neuro for ad operations
Create a Neuro Gem with reusable analysis and reporting instructions, then add the dates and account details for each task.
When AI groups campaigns incorrectly: check names and UTMs
Find campaigns whose names leave product groups or audiences unclear, then use settings, UTMs, and a mapping table to classify them.
How to ask AI when your ROAS numbers don’t match
Why attributed revenue can exceed actual sales, which dates and conversion definitions to align, and how to ask Neuro to investigate a ROAS discrepancy.
Before AI changes your ad budget
How to review AI proposals for budget and bid changes, including approval settings and scheduled execution in Neuro.