# Onfleet: AI search term review for Google Ads

A Claude Code workflow that reviews Google Ads search terms for Onfleet: 561 negative keywords in 9 lists, each checked by hand before going live.

- Industry: B2B SaaS
- Services: Paid media, Marketing operations
- My role: Built the AI workflow and ran paid search
- Audience: Teams that run delivery fleets
- Tools used: Google Ads, Claude Code, HubSpot
- Case study page: https://dilosan.com/work/onfleet/

## The brand

Onfleet makes last-mile delivery software. Teams use it to plan routes, dispatch drivers and track every delivery.

## Why

With AI Max turned on, the search campaigns pulled in 100+ new search terms in a few days, good and bad mixed together. Checking them by hand took hours, so junk terms kept spending money before anyone caught them.

## How

1. I wrote two guides for Claude: an Onfleet knowledge base built from positioning docs and blog posts, and a rule book for negative keywords.
2. I put the guides, a standard template and the search terms report in one folder, then ran the review with one prompt in Claude Code.
3. Claude sorts every term, logs why, and trims brand and competitor names to their root, so one negative keyword covers more ground.
4. I check every suggestion by hand. Only then does it go into the right negative keyword list in Google Ads.

## Step by step

1. **Write the rules down first.** I gave Claude two guides: the Onfleet product knowledge base and a rule book for negative keywords.
2. **Claude sorts every term.** One prompt runs the whole report. Every search term gets a category and a reason in the analysis log.
3. **One master list.** Every negative keyword lives in one master list, grouped by theme, so nothing is added twice.
4. **Checked by hand, then live.** I check every suggestion. Only then does it go into the right shared list in Google Ads.
5. **Paid search keeps closing deals.** Cleaner search terms mean the budget goes to people likely to buy.

## What happened

- 128 companies closed-won from paid search since July 2023
- 561 negative keywords in 9 themed lists
- Low-impression junk terms blocked before they cost a click

## What I learned

- AI is fastest when it has your rules and your product context written down first.
- A human check before anything goes live keeps the speed without the risk.
- I now use the same setup to score ad creatives, test ad copy, format ABM lists and review ads in bulk.

Work by Dilosan Ravindran. Contact: hey@dilosan.com