ChatGPT Ads Context Hints: How to Write Them (With Examples)
Context hints are short descriptions you add to a ChatGPT Ads ad group to explain what you offer, who it helps and when it is useful. ChatGPT uses them, along with your ad and landing page, to decide in which conversations your ad may be relevant.
They are the closest thing ChatGPT Ads has to targeting, and most advertisers write them wrong: they paste keywords, as if it were Google Ads. This guide explains how context hints actually work, gives a simple formula, and shows weak and strong examples from different businesses.
Quick summary: how to write ChatGPT Ads context hints
- Write one clear sentence per hint, about one idea: a product detail, a customer need or a situation.
- Use the what, who, when formula. A hint does not need all three.
- Add context the ad and landing page do not already make clear.
- Keep each ad group to one theme, and start with two or three strong hints.
- Don't paste keyword lists, targeting instructions or a full customer persona.
- Make sure the hint, the ad and the landing page all talk to the same person.
What are context hints in ChatGPT Ads?
In OpenAI's Ads Manager, context hints sit at the ad group level. OpenAI describes them as extra details about your business, product or service that help its systems understand when your ads may be relevant.
Three things they are not:
- Not exact-match keywords. People do not need to use your words for your ad to match.
- Not targeting rules. They cannot limit your ad to a city, a schedule or a type of person.
- Not a guarantee. A hint does not make your ad appear in any specific conversation.
The field is optional. If you leave it empty, ChatGPT still reads your ad and landing page to work out relevance. But a good hint adds the context that a short ad cannot fit.
How does ChatGPT decide when to show your ad?
Ads in ChatGPT appear below the answer, inside the conversation. Each ad shows your name and logo, a title, a short description, an image and a link to your landing page.
According to OpenAI's help center, the ad system looks at several signals together:
- the context and intent of the current conversation;
- your ad's landing page, title and copy;
- your context hints;
- and, when the user has ads personalisation turned on, some signals from their wider ChatGPT use.
So the hint is one signal among several. A strong hint cannot rescue an ad that points to an unrelated page, and the landing page often says more about you than the hint does.
Context hints vs keywords: what is the difference?
The biggest mistake is treating ChatGPT like a search engine. People do not type short queries into ChatGPT. They describe their situation in full sentences and keep talking.
| Google Search Ads | ChatGPT Ads | |
|---|---|---|
| What the user does | Types a short query | Has a conversation |
| Example input | "time tracking app freelancers" | "I'm a freelance designer and I keep forgetting to bill for revisions. How do I track my hours better?" |
| What you give the platform | Keywords and match types | Context hints in plain language |
| What it matches on | The words of the query | The meaning and intent of the conversation |
| Where you restrict delivery | Settings and negative keywords | Campaign settings, not the hints |
A keyword says what someone typed. A context hint says which situation your product solves. That is the shift in thinking.
The formula: what, who and when
OpenAI suggests asking yourself one question: what else should the system know about your product, who it helps and when it is useful?
- What: a feature, service, price, availability or service area.
- Who: the customer's need, preference or circumstance that your offer suits.
- When: the situation or decision in which your offer becomes useful.
A longer version of the same idea, which I find easier for writing several hints:
Who + their problem or situation + the outcome they want + what you offer.
For example, for a home solar installer:
Homeowners with high summer electricity bills who want to cut costs with a rooftop solar system and battery backup.
Who: homeowners. Situation: high summer bills. Outcome: lower costs. Offer: rooftop solar with battery backup. One sentence, one idea.
Weak vs strong context hints: examples by business
The weak versions below are what most people write first. The strong versions describe a real person in a real situation.
| Business | Weak hint | Strong hint |
|---|---|---|
| Meal-prep delivery | healthy meals | Busy professionals who want high-protein lunches delivered ready to eat, without cooking on weeknights |
| IELTS preparation course | IELTS course | Students needing band 7 in IELTS writing for a university or visa application within the next two months |
| Solar installer | solar panels | Homeowners with high summer electricity bills who want rooftop solar with battery backup for power cuts |
| Time tracking app | time tracker, timesheet app, hours app | Freelancers billing several clients by the hour who lose track of small tasks and revisions |
| Wedding photographer | wedding photography | Couples planning a small destination wedding who want candid photos and a same-week preview gallery |
| IT support company | IT services | Small law firms with no in-house IT that need secure email, backups and help when systems go down |
Notice what the strong hints add: a person, a situation and a reason. "Healthy meals" could match thousands of conversations about recipes. The strong version matches the conversation where someone is tired of cooking after work.
How many context hints should you add?
There is no proven ideal number yet, and advertisers do it differently:
- A few focused sentences. Many practitioners start with two or three hints per ad group, each about a different angle of the same theme: one about the problem, one about the outcome, one about a key feature or price.
- Many short phrases. Others go wide. One SaaS team, AirOps, says its working setup uses about 30 hints of 5 to 7 words per ad group, and calls that a starting point, not a proven rule.
- Not one huge block. Reports of very long hint setups, hundreds of words pasted in, describe ads that barely delivered at all.
My advice: start with a small set of specific hints, one idea each. If delivery is low, add more hints that describe other real situations, rather than making the existing ones longer. What every approach agrees on is that the hints in one ad group must stay on one theme.
How to structure ad groups by conversation stage
People come to ChatGPT at different stages of a decision, and each stage needs a different message. Since hints are set per ad group, the clean way to handle this is one ad group per stage or intent.
Here is how that looks for a time tracking app for freelancers:
| Stage | What the person might say in ChatGPT | Context hint for that ad group | Landing page |
|---|---|---|---|
| Problem | "I always feel underpaid for client work, but I'm not sure where my hours go." | Freelancers who suspect they undercharge because small tasks and revisions go untracked | A blog post or page about hidden unbilled hours |
| Solution research | "What's the best way to track time across several client projects?" | Freelancers looking for a simple way to track hours across multiple clients and projects | Features page |
| Comparison | "Is a free time tracker enough or should I pay for one?" | Freelancers comparing free and paid time trackers that can turn tracked hours into invoices | Comparison or pricing page |
| Decision | "Which time tracker exports hours straight into an invoice?" | Freelancers ready to choose a time tracker with one-click invoices and client reports | Free trial sign-up |
Each ad group gets its own ads and, where possible, its own landing page. The person at the problem stage is not ready for a sign-up form; the person at the decision stage does not need a blog post.
When should you create a separate ad group?
Split into a new ad group when the situation needs different messaging or a different landing page. OpenAI's own guidance says the same.
Keep things in one ad group when they are related situations that the same ad and page answer well. And remember the ad group level holds other settings too, such as the maximum bid, so in practice ad groups are mostly where you test different hints and different landing pages against each other.
Common context hint mistakes to avoid
1. Writing a keyword list
"accounting, bookkeeping, tax, invoices, small business" is a list of words, not context. The system matches meaning, not words, so a vague list adds little. Write one natural sentence or a short, specific phrase instead.
Short and specific is not the same as a keyword dump. In one B2B test by the agency Workshop Digital, ad groups with short, specific phrases got a 1.2% CTR, against 0.9% for long conversational hints. Their own reading was that specificity helped, not the keyword format.
2. Giving delivery instructions
"Only show this ad to people in Dubai" or "don't show to students" does nothing. Hints cannot enforce locations, schedules or exclusions. Describing your area is fine as a fact, for example "same-day AC repair in Dubai Marina and JLT". To actually restrict where the ad runs, use the location settings in the campaign.
3. Being too broad
"People who want to save money" fits millions of conversations, and your ad will be weak in almost all of them. Narrow it to the person and situation you serve best.
4. Pasting your whole customer persona
Long paragraphs with age, income, hobbies, job title and pain points are not better hints. Break the useful parts into short, separate hints, each about one idea.
5. A hint that does not match the ad or page
If the hint talks about advanced users but the ad and landing page are for complete beginners, relevance drops and the clicks you do get will not convert. Hint, ad and page should describe the same person.
6. Describing people you cannot really help
Wanting to reach a group does not make your product relevant to them. A hint like "busy parents" on a gaming chair ad only makes sense if your product genuinely solves a parent's problem.
7. Forgetting to update them
Prices, offers, service areas and features change. An old hint that mentions a discount that ended or a city you no longer serve hurts relevance and trust.
How to test and improve your context hints
- Start simple. Clear one-sentence hints in a few focused ad groups, before any complex setup.
- Align the assets. Each ad group's title, copy, image and landing page should match its hints.
- Track clicks properly. Add UTM parameters to your landing page URLs so you can see ChatGPT traffic and conversions in your analytics.
- Compare ad groups. Look at CTR, average CPC and conversions per ad group, not just for the whole campaign.
- Try real conversations. Reporting does not show which conversations your ad appeared in. So open ChatGPT on a free account and have the kind of conversation your customer would have. See which ads appear and what the answers focus on. Testers have noticed ads often show up on the second or third message, not the first, so keep the conversation going.
- Iterate. Split a theme that performs well into its own ad group, and rewrite or remove vague hints that bring weak results.
What does the reporting show?
Ads Manager reporting currently includes impressions, clicks, spend, CTR, average CPC, average CPM and conversions. Impressions and clicks update roughly every 15 minutes, while spend can lag by several hours. There is no report of the conversations or topics that triggered your ads, so judge hints by the results of the ad group they sit in.
Bidding basics for ChatGPT Ads
Context hints decide relevance, but your bid decides whether you win the placement. In Ads Manager you choose:
- Views to pay per impression (CPM);
- Clicks to pay per click (CPC);
- Conversions to optimise toward a conversion event, where available.
You set a maximum bid at the ad group level. For click campaigns, OpenAI recommends starting with a maximum CPC of $3 to $5. The auction weighs relevance as well as price, which is another reason well-written hints and matching landing pages matter.
What have early advertisers found after running ChatGPT ads?
OpenAI publishes no average CTR, CPC or conversion figures for ChatGPT Ads. The best data so far comes from advertisers who shared their own tests. Each is one account at one point in time, so treat them as signals, not benchmarks.
| Who tested | Business type | What they reported |
|---|---|---|
| Workshop Digital | B2B agency | About $2,300 spend, 1.1% CTR, $3.14 CPC, and no qualified leads, while Google Search in the same period brought leads at a higher CPC |
| GravityKit | WordPress software | About $550 spend, 1.7% CTR, $2.57 CPC, no conversions; visitors averaged about 5 seconds of engagement |
| Opascope | E-commerce account | About $60,000 spend over 15 days, $1.72 CPC, 2.35% conversion rate, 1.49x ROAS, with daily ROAS swinging from about 0.2x to 2.9x |
Lessons that come up again and again
- The audience is mostly free users. Ads only show on ChatGPT's Free and Go plans, so many professionals, who pay for Plus or a Business plan, never see them. B2B advertisers felt this most.
- Most traffic is mobile. One tester saw about 70% of paid sessions on phones. Make sure the landing page works well on a small screen.
- Choose the objective carefully. The campaign objective cannot be changed after creation, and ad groups cannot be moved between campaigns.
- Headlines get cut short. Titles and descriptions can be truncated before the official character limit, so put the key words first.
- Ads can pause for review. A brand-safety check can take an ad offline mid-campaign, even when it is fine, so check the review status, not only "active".
- Track on your own side. The in-platform numbers do not show ROAS, and testers found their own analytics recorded fewer clicks than were billed. Use UTM tags and, where possible, both the pixel and server-side conversion tracking.
- Europe works differently. In the EEA and Switzerland, ads are not personalised; they are matched on the conversation, language and general location only.
Are ChatGPT ads worth testing?
ChatGPT Ads is still young, and the results so far are mixed. Some advertisers, especially online stores, see sales at a cost close to their other channels. Many others, especially B2B and lead-generation businesses, get clicks but few leads, and still find Google and Meta cheaper per result.
My view: if your product solves a problem people actually discuss in ChatGPT, run a small, controlled test with tight ad groups and clear hints. Set the budget you are willing to learn with, track conversions properly, and compare the cost per result with your other channels before scaling.
Frequently asked questions
Are context hints required in ChatGPT Ads?
No, they are optional. ChatGPT can still match your ad using the conversation, your ad copy and your landing page. Hints help when your ad and page do not make your best-fit customer or use case clear.
Can I use keywords as context hints?
You can mention important terms, but a list of keywords works poorly. Context hints are not exact-match keywords. Write natural sentences that describe the person and the situation.
Can context hints target a city or country?
No. A hint can mention your service area as a fact, but it cannot restrict delivery. Use the campaign's location settings for that.
How long should a context hint be?
One clear sentence, about one idea. Long paragraphs that mix several ideas are harder to match than several short, focused hints.
Why do my ChatGPT ads get clicks but no leads?
Early advertisers report the same thing, especially in B2B. Common causes: the ad reaches free-plan users who are not your buyers, most clicks are on mobile, the landing page does not match the conversation stage, or conversions are not tracked properly. Tighten the hints, send each ad group to a page for its stage, and check your own analytics, not only Ads Manager.
Can I see which conversations showed my ad?
No. Reporting shows impressions, clicks, spend, CTR, CPC, CPM and conversions, but not the conversations. Compare ad groups by their results, and test conversations yourself in ChatGPT.