Whenever I teach Google Ads basics, I make everyone review the search terms that triggered ads and build a negative‑keyword list.
Spotting negative keywords isn’t hard, but it’s time-consuming and boring. No wonder it was one of the early tasks I’ve tried to hand to ChatGPT. Until last week, though, using AI was more work than doing it manually.
With the latest ChatGPT‑o3, the dullest job in the account finally has a co‑pilot: it can read thousands of search‑term rows and build a comprehensive negative list in minutes.
Why care about negative keywords?
- Every irrelevant click (e.g. “AB University scholarship” or “free online degree”) is budget you never get back
- In Higher Education, you rarely nurture someone from a vague idea to a three‑year programme; Search advertising only works when you focus on people already looking for a place to study
- Google earns money on every click, so it tends to interpret your keywords too broadly. Negative keywords are an important line of defence.
How to do it
1. Gather real data
- Export the last 30–90 days of search‑terms
- Export your current negative list so the model doesn’t reinvent the wheel
2. Ask at a high level
- Give ChatGPT the CSVs and ask it to surface low‑intent words such as free, scholarship, job, application fee waiver, etc.
- Tell it to suggest each word as a broad‑match negative; A single word can block dozens of phrases
Keep the human veto
- Instruct ChatGPT to stop whenever it reaches a borderline word—say review—and wait for your yes / skip before adding it.
- You stay in control while AI does the heavy lifting.
Copy‑and‑paste starter prompt
—
You are a paid‑search analyst for [AB University].
Goal: build a broad‑match negative list that removes low‑intent queries
like “free”, “scholarship”, “jobs”, while keeping study‑intent searches such as “civil engineering bachelor”.
Data attached: search‑terms.csv and negatives.csv.
Return a list in the format:
– negative word or short phrase
– match type (broad unless stated)
– one‑line rationale.
Before adding a word that might block genuine course searches, stop and ask me.
—
In short…
- Export real search‑term data
- Feed it to ChatGPT o3 with a clear, simple prompt
- Approve or skip the suggested negatives
- Upload the list and watch wasted spend drop and conversion rate improve
