How to Write Context Hints for ChatGPT Ads

Every search marketer has spent years learning to think in keywords: match types, negative lists, search term reports, the whole apparatus. ChatGPT advertising discards all of it. In its place is one freeform text box at the ad group level, capped at 280 characters, where you describe the conversations your ad belongs in.

OpenAI calls this a context hint. It is the primary targeting control in the system, and it is the piece most first campaigns get wrong, because keyword instincts actively mislead here.

What a context hint actually does

Your hint is a semantic description that the matching system evaluates against live conversations. It is broad thematic guidance, not an exact-match rule, and OpenAI is explicit that a hint guides matching without guaranteeing delivery.

The auction underneath is relevance-weighted and second-price. Relevance weighting means a more relevant ad can win placement over a higher bid, and the signals include your hint, your ad title and copy, and your landing page content. So the hint is not only targeting. It is an input to whether you win the auction at all and what you pay when you do.

That changes the economics in a way worth exploiting: a precisely written hint can outcompete a larger budget attached to a vague one. Precision here is not just efficiency, it is leverage.

The guidance that follows from the mechanics is to write hints specific enough to describe a real user need, but broad enough to cover the many ways people express that need. Those two requirements pull against each other, and holding both is the entire craft.

The four-part structure

The structure that consistently produces good matching has four components: persona, intent, scope, and disqualifier.

Persona is who the person is. Not a demographic segment, their role and situation. "A security engineer at a mid-market SaaS company." "A founder running a small e-commerce store."

Intent is what they are trying to accomplish, described as they would experience it rather than as your product category. "Trying to add enterprise SSO for a customer that requires it." "Comparing options for reducing false positive alerts."

Scope is the boundary condition that makes your offer relevant: company size, technical constraint, stage of evaluation, budget reality. This is where you narrow to conversations that can convert.

Disqualifier is what makes a conversation not a fit. This one carries more weight than it looks like it should. ChatGPT Ads has no documented negative or exclusion targeting, so a disqualifier written into the hint itself is your only exclusion mechanism. Phrasing like "not for enterprise buyers" or "not for someone looking for a free tool" does real work.

Inside 280 characters you will not always fit all four at full length. Persona and intent are non-negotiable. Scope and disqualifier are what separate a hint that spends efficiently from one that burns budget on conversations that were never going to convert.

Worked examples

Problem-aware stage

Weak: "Authentication, SSO, login security, enterprise SaaS"

That is a keyword list. It names topics, not a person in a situation, and it will match conversations that have nothing to do with your buyer.

Better: "A B2B SaaS founder or engineering lead whose enterprise prospect is asking for SAML SSO before they will sign, who has not built SSO before and is trying to understand the work involved. Not for consumer apps or teams already running an identity platform."

Persona, intent, scope, and disqualifier are all present, and the whole thing describes a moment rather than a category.

Comparison stage

Weak: "Best SSO provider, SSO comparison, Auth0 alternative"

Better: "A technical evaluator comparing enterprise SSO providers for a B2B SaaS product, weighing build versus buy and looking at implementation time and pricing at their scale. Not for someone who has already selected a vendor or needs consumer social login."

The shift is from topic terms to a described evaluation. The hint names what the person is weighing, which is what the matching system can work with.

Purchase-ready stage

Weak: "Buy SSO, SSO pricing, SSO signup"

Better: "An engineering lead ready to implement enterprise SSO who is asking about SAML and OIDC setup, provisioning with SCIM, and how quickly they can ship for a specific customer deadline. Not for early research or general security questions."

The specificity does two jobs: it reaches people close to a decision, and the disqualifier keeps you out of the far larger pool of early-stage conversations that would consume budget without converting.

A consumer example for contrast

Better: "A small business owner running a Shopify store trying to reduce abandoned checkouts, comparing tools on a limited monthly budget. Not for enterprise retailers or agencies managing multiple client stores."

Same structure, different domain.

Six mistakes that waste budget

A keyword list in sentence form. The most common error by a distance. The system matches against a described situation, and a term list describes nothing.

Describing your product instead of their situation. A hint that reads "our platform provides enterprise-grade authentication with SAML, OIDC, and SCIM support" is about you. The match runs against a conversation, which is about them.

Being broad because broad feels safe. A hint that could match half the conversations in your category will match many that cannot convert, and you pay for those. Breadth is dilution here, not reach.

Being so narrow nothing matches. The opposite failure is real. Describe a situation specific enough and few conversations will resemble it, so delivery collapses. Watch impression volume in the first week for this.

Omitting the disqualifier. With no negative targeting available, skipping the disqualifier means having no exclusion mechanism at all.

Using internal or category jargon. If your hint uses vocabulary your buyers never use in conversation, the semantic match degrades. Write the words the buyer says before your team reframes them.

Where good hints come from

The best hints are not invented at a whiteboard. They are recovered from how buyers actually talk.

Mine sales calls and support tickets. The language a person uses when they first describe their problem, before anyone reframes it into category vocabulary, is exactly what belongs in a hint. Map at least ten and ideally twenty real high-intent conversations before writing anything.

Read the questions, not the answers. How prospects phrase the opening question in a demo request, a community thread, or your own inbound is the raw material.

Separate their words from your words. Most marketing teams have thoroughly internalized their own category language. The exercise is recovering what the buyer said before the translation.

Write one hint per audience and intent combination. Hints live at the ad group level, so this is also your scaling structure. Three buyer types across three stages means nine ad groups with nine hints, not one hint trying to serve everyone.

The same discipline shows up in writing for AI citations, where content is retrieved for hidden sub-questions rather than the phrase you targeted. Both reward describing a real situation precisely instead of stuffing terms.

Testing and iterating

Hints are hypotheses. Treat them that way.

Start with three ad groups mapped to decision stage, each with one focused hint, several matched ads, and the most relevant landing page you have. That is enough structure to learn from without building an account you cannot operate. The full campaign setup covers the surrounding structure.

Then let outcomes decide the next move. For each hint the options are to expand it if delivery is good and reach is available, split it if it is matching two distinguishable situations that deserve different creative, loosen it if delivery is too thin, or retire it if it spends without converting.

Judge on cost per qualified outcome, not click-through rate. Independent measurements of the surface run low: SE Ranking recorded a 1.30% click-through rate across roughly 97,000 impressions, and Choice OMG's own June 2026 test came in at 0.65%. A research-stage audience clicks less than a purchase-stage one, and a hint that looks weak on click-through can produce the best downstream pipeline. Killing it on click-through alone is the most common way teams cut the thing that was working.

Why this skill is worth building now

Keyword expertise has a two-decade head start and a deep bench of practitioners. Context hint writing is roughly a year old as a discipline with very little accumulated craft behind it. A team that gets good at it now competes against a field that is mostly guessing.

The structure is also likely portable. The same semantic-input approach is a reasonable bet for other AI ad surfaces as they mature, each tuning its own relevance model over a common style of input. Learning to describe a buyer's situation precisely in a few hundred characters is a skill that probably outlives this particular implementation.

Underneath, it is a discipline good product marketers already have: describing a buyer's situation with precision, in their language, including what they are not. Context hints made that skill directly purchasable as media. Teams with strong positioning have an advantage here that teams with strong keyword operations do not.