ChatGPT Ads work differently from traditional PPC, and the difference starts with how you target.
On Google Ads you usually begin with keywords. On Meta you begin with audiences, interests or behavioural signals. ChatGPT Ads use something closer to conversational targeting: ads are matched to the context and intent of a user’s conversation, with context hints guiding where your product or service is likely to be relevant.
OpenAI describes context hints as broad signals that describe the conversations, topics or needs where your product or service might fit, rather than exact-match keywords. That shifts the job. The mindset moves from buying keywords to testing intent.
What are context hints?
Context hints are descriptions of the conversations where your product or service is likely to be relevant. Instead of matching a search term, they tell the platform what kind of discussion your ad belongs in.
A context hint can cover:
- the product or service you offer
- the problem it solves
- the type of customer likely to need it
- the stage of intent, whether that is research, comparison or purchase
- relevant UK locations
- specific use cases or buying situations
A basic hint might be “people researching [product/service]”. A stronger one might be “people comparing [product/service] providers in London, Manchester, Birmingham or other UK cities”. The second version works harder because it carries the service, the buying intent and the location in one signal.
Why UK advertisers should include locations
If your campaign is set to target the UK only, you should still build locations into your context hints where they are relevant. The two settings do different jobs.
Campaign targeting controls where your ads can serve. Context hints describe the conversations your ads should be eligible to match. One is a boundary, the other is a description of intent.
Consider how different these queries are:
- “What should I know before choosing [product/service]?”
- “Best [product/service] provider in Leeds”
- “Compare [product/service] companies in London”
- “How much does [product/service] cost in the UK?”
Some are broad research. Some are local buying queries. Some are price-led, others comparison-led. They can all be relevant, but they are not the same, which is why location-specific hints are worth testing separately from broader ones.
Test multiple ad groups side by side
Because ChatGPT Ads are still new, your early campaigns should be built to learn rather than to scale. One ad group with a handful of generic hints will not tell you much, and it leaves the platform to do work you should be controlling.
Run several ad groups side by side instead, with each one focused on a single context hint. Until there is a proper experiment feature that splits traffic for you, separate ad groups are the most practical way to compare hints cleanly.
Each ad group should have:
- a clear context hint
- relevant ad copy
- a suitable landing page
- its own UTM tracking
- a controlled budget
- exclusions where needed
Example context hint structure
Here is one way to structure a set of hints for testing.
- Broad research intent tests general demand: “people researching [product/service] options, benefits, pricing or providers”
- Local buyer intent tests location-specific demand: “people looking for [product/service] in London, Manchester, Birmingham, Leeds, Glasgow or other UK locations”
- Service-specific intent tests focused demand: “people comparing providers for [specific product/service]”
- Problem-led intent tests people describing the problem before naming the solution: “people asking how to solve [problem your product/service fixes]”
- Comparison intent tests users weighing up options: “people comparing [product/service] providers, alternatives, costs or reviews”
- Bold positioning intent tests stronger positioning: “people looking for the best [product/service] company in their area”
The point of the test is not to land on the perfect context hint immediately. It is to learn which type of conversation produces the strongest traffic, leads and commercial outcomes.
Example four-ad-group test
A simple version of that test could use four ad groups.
Ad group 1: broad research intent
This targets people in early research mode. Example hints:
- “people researching [product/service] options”
- “people asking whether [product/service] is worth it”
- “people looking for advice before choosing a [product/service] provider”
It tells you whether broad category interest is worth paying for.
Ad group 2: local buyer intent
This targets people looking for a provider in a specific UK location. Example hints:
- “people looking for [product/service] in London”
- “people comparing [product/service] companies in Manchester, Birmingham, Leeds or Bristol”
- “people asking for local [product/service] providers near them in the UK”
It shows whether location-based intent produces stronger enquiries.
Ad group 3: service-specific intent
This targets people who already know the service they want. Example hints:
- “people comparing providers for [specific service]”
- “people asking about pricing, timelines or quality for [specific service]”
- “people looking for companies that specialise in [specific service]”
It tests focused commercial intent.
Ad group 4: bold positioning intent
This tests more assertive positioning. Example hints:
- “people looking for the best [product/service] company in their area”
- “people asking which [product/service] provider is most trusted”
- “people comparing top-rated [product/service] companies in the UK”
It may feel slightly uncomfortable, and it is still worth running. On a new platform, do not assume cautious positioning will always win. Direct, confident language can surface higher-intent demand than you expect.
Budget planning: more ad groups need more budget
The platform has a minimum budget of £15 per day. That gives you roughly £105 per week, or £450 over 30 days. Enough to start testing, but not enough to test everything at once.
The more ad groups you create, the more budget each one needs to produce a clear signal. This is where a lot of advertisers trip up. They build five or six ad groups, spread £15 a day across all of them, and then wonder why the data is weak. Split £15 across six ad groups and each one gets £2.50 a day, which is unlikely to teach you much quickly.
The rule is simple:
- fewer ad groups mean cleaner learning on a small budget
- more ad groups need more budget
- too many ad groups on too little budget give you slow, inconclusive data
Suggested budget structure
For a £15 a day budget, start with two or three ad groups.
- 2 ad groups, £7.50 each. Best for a focused early test.
- 3 ad groups, £5 each. Acceptable, but learning will be slower.
- 4 ad groups, £3.75 each. Only use if you can run the test for longer.
- 5+ ad groups, £3 or less each. Too thin for most early tests.
A sensible first test might run broad research intent at £5 a day, local buyer intent at £5 a day and service-specific intent at £5 a day. Once one ad group starts showing stronger signals, move more budget towards it.
More budget means faster learning
More budget usually means faster learning, because the platform can gather more data across conversation types, context hints, ad variations, click behaviour, landing page engagement and conversion quality.
Current industry commentary suggests ChatGPT Ads may need enough budget to work through a learning period, since the platform needs data across different conversation contexts. AdVenture Media makes a related point: spreading a small budget across too many variants tends to leave results inconclusive.
With more budget behind a test, you can explore more ground:
- more ad groups
- more context hints
- more ad copy variations
- CPC and CPM approaches
- more landing page angles
- more audience exclusions
- faster scaling decisions
On a platform this new, that matters, because you are buying learning as much as traffic.
Smaller budgets can still work over time
A smaller budget can still be useful. It just needs a tighter structure and more patience. At £15 a day, aim for directional learning rather than statistical certainty.
Run over a longer period, a smaller budget can still help you:
- identify weak context hints
- find early performance patterns
- build initial benchmarks
- test message-market fit
- gather lead quality signals
- avoid burning budget too quickly
The mistake is expecting a £15 a day campaign to learn as fast as a larger one, because it will not. Higher budgets buy faster learning, lower budgets buy slower learning, more ad groups always ask for more budget, and poor tracking weakens the learning at any budget.
Use UTMs to track each context hint
UTM tracking is essential here. Each ad group needs a clear UTM structure so you can see which context hint drove the visit, lead or sale. At a minimum, use:
- utm_source=chatgpt
- utm_medium=paid
- utm_campaign=[campaign-name]
- utm_content=[context-hint]
You could use utm_term instead of utm_content. What matters is consistency. Example values might be:
- utm_content=broad-research-intent
- utm_content=local-buyer-intent
- utm_content=service-specific-intent
- utm_content=problem-led-intent
- utm_content=bold-positioning-intent
A full URL would then look like:
?utm_source=chatgpt&utm_medium=paid&utm_campaign=uk_launch&utm_content=local-buyer-intent
With that in place, reporting gets much cleaner. Instead of only seeing that ChatGPT Ads sent traffic, you can see which context hint sent it, and start answering better questions: which hint drove the most clicks, which drove the best engagement, which produced the strongest leads, which brought in poor-quality traffic, and which is worth scaling. Without UTMs, the whole test is weaker.
Use exclude custom audiences
If the aim is to test new demand, use exclude custom audiences where you can. Exclude existing customers, existing leads, contact lists, sales pipeline contacts and previous converters.
This matters most when you are testing context hints. Leave your existing contacts in, and the results can look better than they really are: a hint might appear to be working when the campaign is really being carried by people who already know your brand. The cleaner question is which context hints help you reach relevant new prospects, and to answer it properly you need to exclude the people you already know.
What to measure
Do not judge early ChatGPT Ads performance on clicks alone. Track a fuller picture:
- impressions by ad group
- click-through rate
- cost per click
- landing page engagement
- form submissions
- calls or enquiries
- cost per lead
- lead quality
- conversion rate by context hint
- UTM performance in GA4 or your CRM
- new customer quality after excluding known contacts
The question worth asking is not which ad group got the most clicks, but which context hint produced commercially useful traffic. A broad research ad group might pull more clicks but weaker leads. A local buyer ad group might get fewer clicks but better enquiries. A bold positioning ad group might feel risky and still uncover higher-intent demand. That contrast is the whole point of the test.
Final recommendation
For UK advertisers testing ChatGPT Ads, treat context hints as a strategic testing layer rather than a keyword list. The strongest starting approach is to:
- start with two or three ad groups if you are on the £15 a day minimum
- increase budget when you want to test more ad groups
- keep each ad group focused on one context hint
- include UK locations where local relevance matters
- use utm_content or utm_term to identify each context hint
- exclude existing contacts when testing new customer acquisition
- measure lead quality, not just clicks
- let smaller budgets run for longer, and use larger budgets when you want faster learning
ChatGPT Ads are closer to conversational intent targeting than to another keyword platform. The advertisers who get the most out of them will be the ones who structure their tests properly, give each ad group enough budget to produce a signal, track context hints cleanly, and judge success on commercial quality rather than surface-level traffic.
