# One in fifteen drafts caught before publishing: Sina Shahidi's Google review reply pipeline for five clinics

When Sina Shahidi ran marketing at Pulse Light Clinic, a five-site aesthetics and laser clinic group in London and Manchester, the clinics were mostly not replying to their Google reviews. He built a pipeline to fix that. It pulls new reviews from each Google Business Profile and runs them through a sentiment and topic classifier. It drafts a clinic-specific reply with the OpenAI API and sends the draft to Slack. In Slack a person approves, edits or rejects it. Make.com handles the orchestration. He says the team catches and handles "perhaps one in fifteen drafts before publishing." That is about 1 in 15 drafts stopped by a person. He reports a response rate close to 100% across all five clinics, up from somewhere in the thirty per cent range.

| | |
|---|---|
| **Who** | Sina Shahidi, Director of Marketing at Pulse Light Clinic (2023 to 2026, per his About page), now co-founder of PracticeYield. Writing on [sinashahidi.com](https://sinashahidi.com/insights/ai-google-business-profile-review-responses/) |
| **Problem** | Five Google Business Profiles, each getting somewhere between thirty and a hundred reviews a month, with replies written "when someone remembered" |
| **Stack** | Google Business Profile API, a sentiment and topic classifier (tool not named), OpenAI API, Make.com, Slack |
| **Confidence** | Medium (first-person. Results are self-reported, with no review volumes after launch and no ranking or booking change stated) |
| **Last verified** | 2026-10-27 |

## The problem
Each of the five clinics had its own Google Business Profile. He says each one generated "somewhere between thirty and a hundred reviews a month." Each clinic was supposed to handle its own replies. In practice, he writes, "replies were getting written when someone remembered, in a tone that varied between bored and defensive, and never on a schedule." He also writes that "Negative reviews sat there for weeks without acknowledgement."

He priced out doing the job properly by hand. He estimated "Conservatively, an hour and a half a day across the group." He puts that at "About forty hours a month of someone's time, and that is only if they did nothing else." He built an AI system instead.

## The blueprint (as he describes it)
1. **Pull reviews on a schedule.** The pipeline pulls new reviews from each Google Business Profile through the Google Business Profile API on a schedule.
2. **Classify before drafting.** Each review goes through a sentiment and topic classifier "so we know what is actually being said before we generate a response." The source does not name the classifier.
3. **Draft with clinic context.** The OpenAI API drafts a reply in the tone they want. The draft draws on context for the clinic, the treatment being discussed and the specific points the reviewer raised.
4. **Route every draft to Slack for a human.** A person approves, edits or rejects each draft before it is published. He writes: "Nothing goes live without a person clicking yes."
5. **Orchestrate in Make.com.** He chose Make "because it is faster to wire scenarios in Make than to write a custom backend for something this size."
6. **Set tone per clinic, not per group.** The system has clinic-specific context built in so replies sound like they came from the clinic team, "not a head-office voice." He says this mattered more than he expected.
7. **Spend the care on negative reviews.** Positive reviews can get a slightly varied "thank you, see you soon". Negative reviews need "a careful, specific, non-defensive reply that acknowledges what the reviewer said without admitting things that should not be admitted." He says this is where the AI draft is most useful: "Having a draft to react to is faster than starting from a blank box."
8. **Favour speed.** He writes: "A pretty good reply within an hour is worth more than a perfect reply three days later."
9. **Maintain the prompts.** Since launch, most of his work on the system has been refining the prompts and adding edge cases.

## Stated results
- Response rate "close to 100% across all five clinics, up from somewhere in the thirty per cent range."
- Average response time "inside a few hours."
- Team time of "around five minutes per day per clinic to approve drafts."
- "We catch and handle perhaps one in fifteen drafts before publishing." He says that ratio justifies keeping a human in the loop.
- He says the system "pays for itself many times over in saved time alone." No cost or hours-saved figure is given.
- No review volumes after launch are stated. No change in rankings, bookings or ratings is stated.

## Caveats
- He has since left Pulse Light Clinic. He now does growth work for law firms, medical and dental practices and clinics, and co-founded PracticeYield for UK dental and healthcare practices, so the post doubles as a portfolio piece.
- All figures are his own self-report. The forty-hours-a-month manual cost is his own conservative estimate, not a measured figure.
- "Google rewards profiles that reply quickly" is his claim, with no source given.
- His warning about fully automated reply systems going wrong is a general observation. He does not cite a specific case.
- He says that if he started again he would build the approval interface inside the CRM rather than Slack.
- This is an automated pipeline with LLM drafting and human approval. The source does not describe autonomous agents.
- Handles: LinkedIn [linkedin.com/in/sina-shahidi-73344973](https://www.linkedin.com/in/sina-shahidi-73344973) (verified 2026-10-27: linked from his site; his post announcing the Pulse Light Clinic Director of Marketing role is under it). X: @SinaShahidi (verified 2026-10-27: bio links to truflair.co.uk, the jewellery brand his About page says he co-founded). Bluesky, Threads, Instagram, Mastodon and TikTok are not verified and not tagged.

## Sources
- https://sinashahidi.com/insights/ai-google-business-profile-review-responses/ (primary: first-person build write-up on his own site, 15 May 2026)
- https://sinashahidi.com/about/ (supporting: role, dates and PracticeYield)

---
*Want agents like these working your distribution? [Get It Built](/get-it-built). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*