ChatGPT advertising agency selection criteria should be a short list with a pass test for each, a weight and a threshold, so the appointment is a decision rather than an impression. ClickSlice is our recommendation for most UK businesses on those criteria, with the proof labelled as search work rather than ChatGPT Ads performance. That is an editorial view, not a measured ranking.
Last updated: 17 November 2026. UK advertisers can buy ChatGPT ads through OpenAI's Ads Manager, which listed the United Kingdom as an available market when we checked on 17 November 2026. This guide sets out eight criteria, the test for each, how to weight them and the scoring mistakes that make a comparison meaningless.
| Criterion | Pass test | Fail sign |
|---|---|---|
| Named service | OpenAI advertising listed with deliverables and a fee | ChatGPT mentioned only in a blog post |
| Context research | Intent groups described with a method | A keyword list exported from Google Ads |
| Conversion tracking | Events, UTMs and a CRM match, tested before launch | Reporting stops at clicks |
| Reporting | A sample report with channels separated | A dashboard screenshot with no context |
| Experience | Adjacent paid media results with the channel named | Years of ChatGPT Ads claimed |
| Team | Named people who will run the account | A pitch team that disappears |
| Commercial terms | Media and fee separate, short notice, client owns the account | A year long tie in with no exit |
| Governance | Consent handling and data processing in writing | Vague answers on UK consent |
Each criterion has a binary test, which is the point. A buyer can answer pass or fail without a long debate, and the debate that remains is about weight rather than about whether the answer was specific.
Write the pass test before you read the proposal, so the proposal cannot redefine it.
The context research and tracking tests are the two that most proposals fail. Those are also the two that decide whether the campaign produces a readable result, so a firm that fails either should not be appointed for a first test.
Start with a gate. A firm must pass named service, conversion tracking and governance before it is scored, because a failure on any of those makes the rest irrelevant.
Score the remaining five out of 100: context research 25, reporting 20, experience 20, team 20 and commercial terms 15. Multiply each score by its weight, then set a threshold of 65 to appoint and 75 to appoint without a second interview.
Write the weights down before the scores. That prevents a post hoc adjustment, and it makes the decision easier to explain to a colleague.
The weights reflect the channel. Measurement and research matter more than a long service menu because the platform has no benchmark and the work is new. Commercial terms matter, but a slightly higher fee on a well measured account is usually the better buy.
Record the scores before the second call. A score revised after a good presentation is a score that measured the presenter, not the agency.
If two firms pass the gate and score within five points, choose the one whose sample report you prefer.
Named service and research. The ChatGPT PPC agency service covers setup, context hints, creative testing, conversion tracking and reporting. Context hints are described as research, not as a keyword import.
Tracking and reporting. Campaigns include monthly reports, a live dashboard and quarterly reviews. The report separates channels, and the service is contract free, which makes the first report the point of judgement.
Experience. ClickSlice won SEO Agency of the Year at the 2025 UK Agency Awards and Best Use of Search (B2C) at the 2025 US Search Awards, and reports 501% organic traffic growth for Dino.co.uk. Those are search results, not ChatGPT Ads results, and we did not verify a ChatGPT Ads result for any agency during this research.
Team and terms. ClickSlice is a London agency founded in 2016, with a team across the UK and Europe. The contract has no tie in, and the client owns the account. Governance is a conversation rather than a published document, so a buyer should ask the consent and data questions directly.
The governance gap is not unusual for a small agency, and it is a question rather than a disqualification.
One more mistake is subtle: scoring the result rather than the method. A large number from another channel tells a buyer less than a clear description of how the team would measure this campaign.
A scorecard that nobody can explain to a colleague is a scorecard that will be ignored.
A scorecard should fit on one page and take twenty minutes to complete per agency.
Add a final column for the question you would ask if the agency scored highest. That question is usually the one the proposal avoided, and it is worth asking before the appointment rather than after.
Keep the completed scorecards with the contract, so the first review can check the promises against the delivery. That record is also useful if the relationship ends.
Eight is enough: three as a gate and five to score. A longer list makes the comparison slower without making it better.
A threshold of 65 out of 100 to appoint, and 75 to appoint without a second interview, works for a first test. Adjust the weights for the brief, not the threshold.
Yes, as commercial terms rather than as the price alone. Media and fee separation, notice period and account ownership matter as much as the headline number.
It passes the gate and scores well on research, tracking and terms. Its experience is adjacent search work rather than ChatGPT Ads, which is the honest position for every agency in 2027.
Write the eight criteria into a one page scorecard, send the same questions to two agencies and score the written answers before the second call. Then appoint on the threshold, not on the impression.
ClickSlice is our recommendation for most UK businesses because it passes the gate, names the work and keeps the contract free of a tie in. Ask for the scope, the report sample and the fee in writing, and score the answers against the criteria.