# Ask for builder reviews at a verified buyer milestone

An AI-drafted review request starts with a verified buyer milestone, the right recipient, recorded contact permission and an approved review destination. The same eligibility rule applies to every buyer. Service concerns go to their owner on a separate track. A person checks the neutral wording, the link and the platform's current guidance before any request goes out.

## When is a review request ready to draft?

When a verified milestone exists, not when someone has a good feeling. A productive design meeting, an approved [handoff](/process/handoff/), move-in or a resolved service interaction can each be the moment, depending on your process. A person confirms the milestone happened. The same eligibility rule applies to every buyer who reached the milestone.

Open punch work, disputed selections, warranty concerns and unanswered messages go to the service owner. Keep that work out of review eligibility. Apply the same milestone and contact criteria to every buyer, including buyers with open concerns. The agent never uses a service record as a proxy for expected sentiment.

The [AI for builders hub](/articles/ai-for-builders/) shows how this workflow sits beside reviewed drafting and record controls.

### What evidence supports the trigger?

| Trigger input | Agent check | Human confirmation |
| --- | --- | --- |
| Project milestone | Find the approved status in the project record | Confirm the milestone is complete and suitable |
| Buyer identity | Match the approved contact to the project | Confirm recipient and relationship |
| Communication permission | Check the recorded channel status | Confirm the planned outreach is appropriate |
| Open service items | Flag records for the service queue | Assign a service owner. Do not drop the buyer from the request milestone |
| Review destination | Insert the approved link or instructions | Confirm the destination and current platform guidance |

## What does the request say?

Keep it short and specific to the relationship. Name the verified milestone, ask for an honest account and give the approved path to respond. Do not supply adjectives, suggest a rating or write a review the buyer can paste. The buyer's words stay the buyer's words.

The request can remind the buyer which part of the process just closed: design, construction handoff, service follow-up. It does not claim the whole experience was positive. Neutral language keeps the request credible and leaves room for real feedback.

### What does the agent never infer?

Satisfaction. Not from a friendly email, a signed document, a completed closing or silence after a warranty visit. Those are events, not sentiment. The agent may draft from a recorded buyer comment when the source is clear. It still does not turn that comment into a public endorsement.

## How do you build a reviewed request workflow?

1. **Define eligible milestones.** The project events that may prompt a request and the evidence each one needs.

2. **List neutral stop conditions.** Unknown recipient, missing contact permission, unverified milestone, duplicate request. Service work routes separately.

3. **Approve the template.** Sender, tone, review destination and neutral request language.

4. **Prepare the draft.** The agent inserts verified buyer, project and milestone facts only.

5. **Check consistent eligibility.** Same milestone and contact criteria for every buyer. Open issues go to a service owner on their own track.

6. **Approve the send.** Recipient, link and platform instructions, checked at the moment of use.

7. **Record the action.** Request date and outcome. No fabricated review status.

The [AI operations stage](/process/ai-operations/) defines the agent's access, the human approval and the failure rule. An uncertain project record stops the workflow for review.

The [human review checklist for AI](/articles/ai-for-builders/human-review-checklist-for-ai/) covers the trigger, the recipient, open issues, the link and the final send decision.

## How do you handle platform and policy questions?

Check review requests, reminders and reuse against the destination platform's current official guidance. For a [Google Business Profile](/glossary/google-business-profile/), that means Google's own published rules, read at the time of use. Check outreach and data-use obligations with qualified legal counsel and the controlling sources. Save the guidance you reviewed next to the approved template. The agent never states remembered policy as current.

Never design a workflow that sends only to buyers predicted to be happy. Keep the same public review destination open to every eligible buyer. Audit skipped requests against the written neutral criteria, including buyers with open service records. The companion article on [review policy compliance](/articles/reviews-and-reputation/review-policy-compliance/) lists what to verify before launch.

### Can AI answer the review once it appears?

It can draft a response from the public review and approved project context. A person verifies identity, privacy boundaries, service facts and tone before posting. Never reveal contract, warranty, selection or personal details to win a public argument.

## Where does this workflow meet aftercare?

Keep review-request and service records linked and separate. Every service concern has an owner and a current status, whatever the review says. When a buyer raises an issue, route it the same day. Apply the same reminder limit and opt-out rule to every recipient. Sentiment never changes the sequence.

The [builder reviews program guide](/guides/builder-reviews-program/) covers milestones, response ownership and builder-controlled records. The article on [AI for warranty requests](/articles/ai-for-builders/ai-for-warranty-requests/) covers intake and clarification for service work.

At AgAI Labs, our Claude and Codex AI agents research current source material, draft the message and check required fields. People review every output and stay responsible for service decisions, sending and public responses.

### Which review request questions come up with AI?

### Can AI decide which buyers are happy?

No. Use verified milestones and one eligibility rule for everyone. Never filter requests by predicted satisfaction or by open service concerns.

### Should we write sample review text for buyers?

No. Ask for an honest review in the buyer's own words. Do not script praise, supply sample text or suggest a rating.

### What if a buyer replies with a service issue?

Route it to the service owner. Apply the same reminder and opt-out rules to that buyer as to every other review recipient.

### Can the agent post a response automatically?

No. AI review responses start as drafts. A person checks identity, project facts, privacy, tone and unresolved service work before anything goes public.