ChatGPT Ads Context Hints: The Complete Guide (2026)

Context hints are the short descriptions you give ChatGPT Ads to explain when your product or service is relevant. They describe what you offer, who it helps and the situations where someone needs it. They are not keywords, and they don't decide exactly which conversations show your ad.

I spent more than ten years in Amazon advertising and built PPC Entourage, an Amazon ad analytics tool acquired by Carbon6 in 2021. The lesson I learned over and over was that the accounts that won weren't the ones with the longest keyword lists. They were the ones that understood exactly which customer need each dollar was buying. Context hints push you toward that kind of thinking from day one.

This guide covers what context hints are, how to write and organize them, how many you can use, and how to tell which ones are working.

Context hints are attached to an ad group. OpenAI describes them as information that helps its system understand when your ads may be relevant, and asks that they add detail your ads and landing pages don't already make clear.

Take a local plumbing company. On Google, it might bid on searches like "emergency plumber near me" or "water heater repair." In ChatGPT Ads, it describes the need instead:

Emergency residential plumbing repairs for homeowners dealing with burst pipes or sudden water leaks.

Water heater repair and replacement for homeowners with no hot water or inconsistent hot water.

Each one ties a real customer situation to a real service. That's the whole idea.

This is the biggest shift for anyone coming from Google Ads.

OpenAI's API reference calls context hints "descriptions or keywords," while its advertiser help center says they are "not exact-match keywords." Both are true: a hint can contain words people use, but it isn't a Google-style keyword target. Your job is to explain relevance, not to dictate delivery.

OpenAI asks for specific, accurate, natural-language descriptions. I use three questions. A hint doesn't need to answer all three, but the best ones usually answer two.

1. What exactly do you offer? Not "plumbing" but "residential water heater repair and replacement."
2. Who is it for? Not "accounting software" but "bookkeeping software for small retailers selling online and in store."
3. When does the need show up? Not "roof repair" but "roof inspection and repair after storm damage or a new leak."

One rule above the rest: only describe what's true. If you don't offer same-day service, don't write "same-day." If you only serve one county, say so in your location targeting, not by stuffing city names into hints.

These examples show the writing pattern. They're made up for this guide and haven't been tested for performance.

Up to 2,000, according to OpenAI's campaign targeting documentation. That's a ceiling, not a goal. OpenAI itself says you don't need to reach the maximum. A dozen accurate hints beat hundreds of near-duplicates.

One practical detail from the same docs: when you update an ad group's hints, the new list replaces the old one completely. Anything you leave out is removed, so keep a master copy of every ad group's hints outside Ads Manager.

ChatGPT Ads follows a familiar structure: campaign → ad group → ads. Context hints live on the ad group. OpenAI recommends building each ad group around one product, service or customer need, and splitting groups when they need different messaging or landing pages.

For a plumbing company, the standard setup looks like this:

Each group gets its own ads and the landing page that fits it. This is a good starting point, and it's how most advertisers should launch.

Grouping has a blind spot. When an ad group with 15 hints books a call, you can't tell which customer need drove it. You know the group worked, not why.

The Big Hint Campaign takes the opposite approach for testing. It gives each important hint its own ad group, with the same creative and landing page across groups. Because everything else stays equal, each ad group's results show how that one customer need performs, measured in cost per booked call.

The trade-offs are real:

• It needs budget. Each group needs enough spend to get meaningful volume.
• Small numbers mislead. A group with two booked calls isn't proof of anything yet.
• The hint isn't the only factor. Delivery also depends on bids, creative, targeting and OpenAI's own systems, so results show how each ad group performed, not exactly why an ad was shown.

So treat it as a test, not a permanent structure. Run it, find the customer needs that book calls cheaply, then put more budget behind them. Big Hint builds this campaign for you and compares every hint side by side.

Your best hints are already inside the business. They're the questions customers ask before they buy. Start with five places:

1. Sales calls: what prospects ask before they commit
2. Customer service: the problems that make people reach out
3. Your CRM: which kinds of leads turn into paying customers
4. Your offer: what genuinely makes it right for a specific need
5. Objections: the situations that decide whether someone picks you

Read through 100 recent inquiries at a plumbing company and the themes jump out: emergencies, water heaters, recurring clogs. Each one is a candidate hint, and a candidate ad group.

Keep customer data private while you do this. Summarize the patterns yourself. Don't paste real customer messages, names or contact details into outside tools.

Writing hints is the easy part. The real question is which ones bring you business. Look past clicks to what happens next:

• Spend, impressions, clicks, click-through rate and cost per click
• Conversions and cost per acquisition, once tracking is set up
• Qualified leads and booked calls from your CRM
• Revenue, where you can attribute it

For a lead-generation business, cost per booked call usually tells you more than cost per click. Here's a made-up example of two ad groups with the same budget:

Emergency repairs books calls at half the cost, so it's worth testing more budget there. It doesn't prove the hint alone made the difference: check lead quality, how many calls became jobs, and whether the ads and bids were the same.

The question isn't "which hint got the most clicks?" It's "which customer need is making us money?"

1. Treating hints like exact-match keywords. A hint doesn't mean "show my ad whenever someone says this." Write for relevance.
2. Being vague. "Home services" tells the system almost nothing. Name the service and the need.
3. Claiming what you don't offer. No invented availability, certifications, locations or deals.
4. Using hints as location rules. A hint can mention where you work, but it doesn't restrict delivery. Use the campaign's location targeting for that.
5. Overwriting your list by accident. Saving an ad group's hints replaces the whole list. Keep a master copy and paste the full list every time.
6. Changing everything at once. If you change hints, ads, bids and landing pages together, you won't know what moved the numbers. Change one thing and write down what and why.

1. List your most valuable products, services and customer needs.
2. Pull recurring questions from sales calls, service requests and your CRM.
3. Write specific, accurate hints for each need.
4. Group related hints into ad groups, or give each key hint its own group to test it.
5. Write ads and pick landing pages that match each group.
6. Launch with conversion tracking and CRM results connected.
7. Compare ad groups on cost per booked call, not just clicks.
8. Shift budget toward what works, log every change, and repeat.

Big Hint is ChatGPT Ads software for agencies and brands. It builds a campaign around your context hints, connects each one to the leads and booked calls in your CRM, and shows you what each booked call costs and what to change next.

• OpenAI Help Center: Create ad groups and context hints for ChatGPT Ads
• OpenAI Help Center: Write context hints for ChatGPT Ads
• OpenAI Developers: Campaign targeting
• OpenAI Developers: Ad groups API reference

Last reviewed October 27, 2026. ChatGPT Ads changes often; we update this guide when OpenAI's documentation does.