Keep the AI Conversation Going

Why the follow up matters

OpenAI recommends reviewing an initial response and refining the request with clearer wording, additional context, or a narrower task. Anthropic similarly encourages follow-up instructions, clarification, and requests for revision. These are practical recommendations for directing the work; they do not promise that every revision will be accurate. [1, 2]

Continuing a focused conversation lets you build on the material already under discussion. OpenAI's prompting guidance specifically describes adding a missing source, correcting direction, or changing the level of detail without starting over. [3]

The useful follow-up names what should change and why. A profile might need stronger evidence. A priority list might need a realistic limit for the week. A meeting summary might need to separate what you said from what people actually said in response.

Here are four examples. Each assumes that you have already supplied the relevant source material and received an initial answer. Use an organization-approved tool for donor information and internal notes.

1 Move individual research toward a decision

An initial profile may organize a person's career, giving history, and affiliations. The next question is whether that information gives your organization a documented reason to explore a relationship.

A follow-up can focus the research on mission connection and identify the missing information that would change your next step.

Follow-up prompt

Using the profile and sources above, identify up to three documented connections between this person and our mission of [mission]. For each, show the source and date, distinguish confirmed facts from your interpretation, and flag identity mismatches or missing evidence. Finish with the one unanswered question that would most affect our qualification decision. Do not infer willingness to give from wealth.

This gives the researcher something specific to verify before recommending outreach. A plausible connection still needs support from the underlying source.

2 Turn a large list into weekly priorities

Suppose you have already asked AI to review a spreadsheet of 1,000 prospects. A broad ranking may help you understand the pool, but the immediate decision is where your team should spend its time this week.

For example, a documented commitment to reconnect this week may deserve attention even when that prospect is not at the top of the capacity list. Your follow-up should apply the organization's existing rules and make the reason for attention visible.

Follow-up prompt

Using the same spreadsheet, our supplied prioritization rules, and the week of [dates], identify up to 10 people who need attention this week. Consider due or overdue commitments, recent meaningful engagement, open opportunities, and gaps in follow-up. Exclude completed or canceled actions, and account for follow-up already scheduled. Show prospect ID, the documented reason for attention now, one suggested next action, and the supporting fields and dates. Preserve IDs exactly, report how many records you actually evaluated, and flag missing information.

Use a tool that can analyze the actual spreadsheet, and check its reported record count and recommendations against the file. Missing contact notes should remain an information gap; they do not establish that no contact occurred.

3 Prepare a more useful donor conversation

After receiving a donor briefing, you can use the same conversation to prepare for the meeting itself. The aim is to turn documented interests and previous comments into talking points that leave room for the donor to tell you more.

Follow-up prompt

Using the donor briefing and previous contact notes above, develop three talking points and one open-ended question for each for our next meeting about [purpose]. Tie each talking point to a documented interest or prior comment and show the supporting note. Label proposed discussion ideas as suggestions. Do not invent personal interests, motivations, or readiness to make a gift.

For example, a donor's previous question about scholarship access provides a grounded topic to revisit. It does not establish that the donor wants to fund a particular program. Use the preparation to guide your listening.

4 Understand the response to your leadership

An executive may start by asking AI to summarize meeting notes. A useful next step is to examine the discussion more closely: the main points the executive raised, the feedback recorded from others, and the questions that remain open.

Return to the original notes or transcript for this step. The first summary may have left out details you now need.

Follow-up prompt

Return to my original notes or transcript and use the summary as a guide. In a concise table, show each key topic, the point I made, the feedback others actually gave, and any decision or unresolved question. Attribute comments only when the speaker is identified, and include the supporting note or timestamp. Where feedback is absent, write 'not documented'; do not interpret silence as agreement.

This can help you identify a concern that needs a response or a decision that needs clarification. Review it against the notes before treating it as a record of what the group agreed to.

Keep the conversation grounded

Continuing a conversation is useful while its context remains relevant and accurate. A 2025 study by Microsoft and Salesforce researchers found that models could become anchored to early mistakes in simulated conversations where requirements emerged over several turns. The study examined specific tasks and models, rather than nonprofit fundraising outcomes, but it supports being deliberate about when to reset. [4]

If a thread keeps repeating an incorrect assumption, consolidate the verified facts and current instructions into a clean brief and start again. Within a useful thread, add evidence and corrections as you go. Ask for support for a recommendation, and check that support yourself.

Sources

1. OpenAI — Prompt engineering best practices for ChatGPT
Guidance on clear instructions, context, and iterative refinement.

2. Anthropic — My prompt isn't giving me a helpful answer
March 16, 2026. Guidance on follow-up instructions, clarification, and revision.

3. OpenAI — Prompting
Guidance on improving an answer through follow-up messages and checking the result.

4. Laban, Hayashi, Zhou, and Neville — LLMs Get Lost In Multi-Turn Conversation
May 9, 2025. Study of simulated conversations; supports the qualification about persistent errors and consolidating instructions.