A context hint is the targeting primitive of ChatGPT Ads, replacing the keyword in conversational AI advertising. It is a 1-to-2-sentence plain-language description of a user moment - the persona, question, outcome, or stack comparison the advertiser wants to reach inside a ChatGPT conversation. The platform's semantic matcher reads the hint as an intent signal, not as a string to literal-match. Where a Google Ads keyword names a search query, a context hint names a buyer.
- One hint covers many phrasings. The matcher works on semantic similarity, so the same hint reaches users who express the same intent in different words.
- Five canonical patterns. Every working hint classifies into one of five shapes: Persona + Intent, Question, Topic + Disqualifier, Outcome, or Stack Comparison.
- Length: 1-2 sentences, under 280 characters. Long enough to encode persona and scope. Short enough to keep semantic precision.
- Live in the self-serve dashboard since May 5, 2026. Open to every US advertiser at ads.openai.com, no minimum spend.
What is a context hint?
A context hint is a sentence an advertiser writes to describe the user moment they want their ad to appear inside. It is the input that ChatGPT Ads' matcher uses to decide which active conversations qualify as ad surfaces for that advertiser. The hint encodes who the user is, what they are trying to accomplish, and any context that scopes the conversation. The matcher then reads the embedding of inbound conversations and serves the placement when an inbound conversation aligns with the hint's intent.
The simplest way to understand a context hint is by analogy to the thing it replaced. A Google Ads keyword like best CRM for sales teams is a string the matcher tries to align with words a user typed. A context hint like Marketing leaders comparing CRM platforms for a 50-person team is a description of a buyer the matcher tries to align with words a user is currently saying inside a ChatGPT conversation. The keyword names a query. The hint names a person at a moment.
That shift - from string-matching to person-matching - is the entire point. ChatGPT users do not type four-word search queries. They type sentences, paragraphs, and follow-ups. The conversation is the unit of attention, not the keyword. A targeting primitive that matches the user's mental frame outperforms one that tries to match the user's literal text. Context hints are how OpenAI made that primitive available to advertisers.
A keyword names a query. A context hint names a person at a moment.
Where context hints live in the ChatGPT Ads dashboard
Context hints live inside the self-serve ads.openai.com dashboard, which OpenAI opened to every US advertiser on May 5, 2026. The dashboard exposes context hints as a free-text targeting field at the ad-group level. Advertisers describe the user moment in 1 to 2 sentences. There is no keyword-list import, no negative-keyword sheet, no exact-match-vs-broad-match toggle. There is one field, and the field accepts plain English.
The placement of the field inside the campaign builder follows the standard wizard order: campaign objective (Reach, Clicks, or Conversions), then ad group, then ad creative. The context hint sits at the ad-group layer because that is where targeting belongs in the platform's data model. One ad group, one hint. Multiple ad groups per campaign let an advertiser run parallel hints against parallel buyer segments without cross-contaminating the matcher.
Two surrounding mechanics matter for understanding how the field works. First, the matcher accepts a character limit of roughly 280. Hints longer than that drift across multiple buyer scenarios and the matcher's precision drops measurably. Second, the field is paired with the campaign objective: a Reach campaign uses the hint to optimize for impressions inside qualifying conversations, while a Clicks campaign uses the same hint as the targeting input but bids against engagement. The Conversions objective reaches early access on June 5, 2026 via conversion-optimized campaigns for accounts that set up the pixel or Conversions API by June 1, with OpenAI's conversion pixel and a server-side Conversions API having launched broadly on May 5, 2026.
For advertisers coming from Google Ads, the surface area is striking for what is missing. There are no audience lists. There are no demographic toggles. There is no keyword bidding screen. The field for the context hint and the field for the bid amount are the only two inputs that determine reach and price. Everything else in the wizard is creative and reporting. OpenAI's official ads basics documentation walks through the full wizard flow.
How context hints are priced
A context hint sets targeting; the bid sets price. ChatGPT Ads runs on a cost-per-click model with category-level bid floors surfaced in the self-serve dashboard, in the $3 to $5 range as of April 27, 2026 per our walkthrough of the dashboard. The context hint and the bid amount are the only two inputs that move reach and cost: a more precise hint reaches fewer, better-qualified conversations, while a higher bid clears the category floor and wins more of the eligible placements.
Two rules shape planning. The bid floor is a hard gate: bidding below the category floor means the placement does not deliver, no matter how strong the hint. And the objective shapes how the bid is spent: a Reach campaign optimizes for impressions in qualifying conversations, a Clicks campaign bids against engagement, and a Conversions campaign (via conversion-optimized campaigns) optimizes toward pixel or Conversions-API events. For the full cost breakdown, see our ChatGPT Ads cost guide.
Why context hints replaced keywords
The replacement is not a UX choice. It is a consequence of how ChatGPT works. Google's keyword auction was built for a search box where the user types a 2-to-5-word query and an algorithm matches that string against advertiser bids. ChatGPT does not have a search box. It has a conversation field where the user types a sentence, a paragraph, or a back-and-forth thread. The unit of attention is the conversation, not the query.
Inside that conversation, OpenAI's matcher uses semantic embedding rather than lexical comparison. An embedding is a numerical representation of meaning, computed by the same family of models that power the rest of the ChatGPT experience. When a user types "I'm evaluating Salesforce vs Blaze CRM for our 50-person sales team. Which one scales better past Series B?", the embedding of that conversation lands in a high-dimensional space close to other conversations expressing similar intent, regardless of the exact words used.
A context hint is the same kind of object: an embedding the matcher computes once when the advertiser saves the hint, then compares against incoming conversation embeddings in real time. The closer the two embeddings sit in space, the more eligible the placement is for that conversation. There is no string to literal-match against. The Google Ads keyword field would have no role to play in this system, because there is no point in the data flow where two strings need to overlap.
The practical consequence for advertisers is that the breadth-first instinct that worked in Google Ads is the wrong instinct in ChatGPT Ads. Adding 200 long-tail keywords to a Google Ads campaign generally improves coverage. Adding 200 context hints to a ChatGPT Ads ad group dilutes match precision because the embeddings collide and overlap. The discipline inverts. Fewer, more precise hints outperform more, less precise hints. That inversion is the single biggest mental adjustment a Google Ads marketer makes when moving to conversational ad targeting, and one of five strategic shifts that follow from the targeting-primitive change. The campaign setup walkthrough covers the wizard mechanics for marketers running through this transition for the first time.
The 5 Context Hint Patterns
Every working context hint classifies into one of five shapes. The 5 Context Hint Patterns are the canonical translation taxonomy for converting Google Ads keywords into ChatGPT Ads context hints, and the reference framework we use across our migration methodology, our free converter tool, and the glossary entry. The framework was observed across thousands of B2B and SMB campaigns during prompt design and is now the standard definitional set.
1. Persona + Intent
Persona + Intent is the workhorse pattern. The hint names WHO the user is plus WHAT they are trying to accomplish, in that order. It is the right pattern when the underlying buyer is well-defined by role, company size, or industry, and when the action they take inside the conversation is comparison, evaluation, or shortlisting. Most BOFU keywords (the ones that drive the highest conversions in Google Ads) translate directly into this pattern.
2. Question
Question hints mirror the WH-question or how-to phrasing the user actively asks. It is the right pattern when the underlying buyer journey is a how-to query rather than a comparison. The matcher reads "users asking how to set up multi-channel attribution" as a description of a moment, not a literal-match string. Most informational TOFU keywords translate into this pattern, though the conversion rate is typically lower than Persona + Intent because the user has not yet shortlisted vendors.
3. Topic + Disqualifier
Topic + Disqualifier hints add an explicit "not X" filter to a topic, which tells the matcher where the placement should not appear. It is the right pattern when adjacent intents drain spend - the most common case being implementation or troubleshooting traffic for a product when the campaign is for evaluation traffic. Disqualifiers are how the platform's targeting primitive handles the gap left by negative keywords in Google Ads, but they are richer because they describe the scope of the desired moment, not just the words to exclude.
4. Outcome
Outcome hints describe the specific result the user is trying to reach. "B2B teams trying to cut sales cycle from 90 to 60 days" is an outcome hint. The matcher aligns with conversations where the user expresses the same goal, even if they never name a product category. This pattern is unusually strong for high-consideration B2B because it bypasses the category-naming layer entirely - the user does not have to know they need a CRM to land on a CRM-categorized hint.
5. Stack Comparison
Stack Comparison hints describe a migration from one named tool or stack to another. "HubSpot users considering switching to Salesforce" is a stack-comparison hint. It is the right pattern for displacement plays - winning customers from a known competitor or completing a tool migration the prospect has already started. The pattern reaches conversations where the user names both stacks even when they do not type the word "switch" or "migrate."
| Pattern | Definition | Blaze CRM example |
|---|---|---|
| Persona + Intent | Names WHO the user is plus WHAT they want to accomplish. | "Marketing leaders comparing CRM platforms for a 50-person team" |
| Question | Mirrors the WH-question or how-to phrasing the user actively asks. | "Users asking how to set up multi-channel attribution" |
| Topic + Disqualifier | Names a topic plus an explicit "not X" scope filter. | "CRM evaluation, not implementation help" |
| Outcome | Describes the specific result the user is trying to reach. | "B2B teams trying to cut sales cycle from 90 to 60 days" |
| Stack Comparison | Describes a migration from one named tool or stack to another. | "HubSpot users considering switching to Salesforce" |
Hints that do not classify into any of the five are signals the underlying buyer scenario is unlikely to translate to ChatGPT Ads at all. The discipline is to reject them rather than force-fit. A keyword like "marketing software" with no persona, no intent, no question, no disqualifier, no outcome, and no stack reference does not have a coherent buyer behind it. The pattern set is the filter that separates targeting opportunities from spend traps. For 2 worked examples per pattern, scored against the launch-gate rubric, see the 10 best hint examples companion post.
Anatomy of a good context hint
A good context hint has up to four components: persona, intent, scope, and (optionally) a disqualifier. Persona names WHO. Intent names WHAT they want. Scope narrows the qualifying context (company size, role, vertical). Disqualifier names WHERE the placement should not appear. The first three are present in most working hints; the disqualifier is added when adjacent intents would drain spend.
The Blaze CRM example we have been using - "Marketing leaders comparing CRM platforms for a 50-person team" - has persona ("Marketing leaders"), intent ("comparing CRM platforms"), and scope ("for a 50-person team"). It does not yet have a disqualifier, which means the matcher will eligible the placement for users actively comparing CRMs but also for users in the implementation phase asking how to set up their existing CRM. If the campaign is for evaluation traffic only, that drift wastes spend. Adding a disqualifier - "not implementation help" - tightens the qualifying context significantly. The complete 4-component hint reads:
Two discipline points govern length. First, the hint is 1 to 2 sentences. Anything shorter loses one of the four components and the matcher loses signal. Anything longer drifts across multiple buyer scenarios and the matcher's precision drops measurably. Second, the practical character cap is around 280. The exact ceiling is not officially documented but is consistent with what the platform's UI accepts and what behavioral testing confirms; treat it as a working ceiling, not a hard rule.
A third discipline applies to the language register. The hint reads like a sentence a strategist might write to brief a copywriter on the buyer scenario, not like a search query a marketer would target. Concrete persona language ("CTOs at Series B fintech startups") outperforms generic role labels ("decision-makers"). Concrete intent verbs ("comparing," "evaluating," "switching") outperform abstract goals ("interested in"). Concrete scope numbers ("50-person team," "$10M revenue," "3-to-5-year horizon") outperform vague qualifiers ("growing," "mid-sized," "soon"). The same register discipline applies to writing the ad copy itself.
A context hint reads like a strategist's brief. Not like a search query.
The fastest way to internalize the anatomy is to take 5 of your existing Google Ads keywords and translate them by hand. Group them by which of the 5 Patterns they fit, then write the hint with the four components labelled. Part 3 of this cluster, the 5-Step Hint Method, runs the procedure end-to-end with a Hint-Quality Scorecard for grading the result before launch. Automating the classification is fast; the manual exercise is what builds the muscle memory that the automation cannot.
Paste 5 Google Ads keywords. Our converter classifies each into one of the 5 Patterns and outputs the matching context hint. No signup. Try the free Google Ads to ChatGPT Ads Converter.
What context hints are NOT
Five things that look like context hints but are not, framed as definitional clarification rather than common mistakes. If you find yourself writing one of these, the underlying buyer scenario probably needs a different translation, not a tighter hint.
A generic topic
"Marketing software" is not a context hint. It has no persona, no intent, no question, no scope. The matcher treats it as a topic landmark and serves the placement against any conversation that gets within range of that topic, including conversations from current customers, students researching a class project, or people writing about marketing software for completely unrelated reasons. Topics without people behind them are not targeting; they are noise generators.
A keyword string copy-pasted from Google Ads
"Best CRM for sales teams" is not a context hint either, even though it has a topic and an implied audience. The Google Ads keyword is written for a search-query matcher; the conversation matcher reads it as a fragment, not as a description. The fix is to expand it into the persona-and-intent format the conversation matcher expects: "Marketing leaders comparing CRM platforms for a 50-person team." Same scenario, different format.
A long brief
A 5-sentence paragraph describing the ideal customer profile, the use case, the disqualifying scenarios, the budget thresholds, and the buying-committee composition is not a context hint. It is a buyer-persona document. The matcher loses precision when the hint covers too many scenarios. If the brief reads like a paragraph, split it into 2 or 3 ad groups, each with its own focused hint.
A demographic
"CMOs aged 35-55 in tech companies" is not a context hint, because demographics are not part of the targeting model. ChatGPT Ads matches on conversational intent, not on user identity attributes. The hint that captures the same buyer is "marketing leaders evaluating new tools for their stack" - which describes the moment, not the person's birth year.
A literal user query
"How do I pick a CRM?" is not a context hint, because the hint is a description of the user, not a transcript of what the user types. The matcher reads "users asking how to pick a CRM" as a description of a moment and aligns to many phrasings of the same question. Writing the hint in the user's voice strips out the descriptive layer the matcher needs to generalize.
What's still unclear about context hints
Context hints are 12 weeks old as a public targeting primitive. Some mechanics are documented, some are inferable from behavioral testing, and some are open questions the platform has not addressed yet. Four worth flagging for any advertiser planning to spend on the system in 2026.
- Optimal hint count per ad group. OpenAI does not publish a recommendation. Behavioral testing suggests 1 hint per ad group performs better than 3 or 5, but the data is thin and the platform may evolve as advertisers experiment.
- Multilingual matching. The matcher works on semantic embeddings, which are inherently multilingual at the model level. Whether a hint written in English reaches Spanish-language conversations is not officially documented, and behavioral testing on this is early.
- Hint-to-hint conflict resolution. When two advertisers' hints both qualify the same conversation, the auction mechanism is described as price-based, but the tiebreaker between equally-priced hints is opaque. This matters for displacement campaigns where multiple advertisers target the same Stack Comparison moment.
- Brand-name handling. Does naming a competitor brand in the hint ("HubSpot users considering Salesforce") create eligibility for that competitor's brand-only conversations, or does the matcher require the comparison structure to be present? Anecdotal reports vary; OpenAI has not commented.
The platform is moving fast. OpenAI's May 5 expansion announcement opened the Ads Manager to every US advertiser, the third major access change in twelve weeks (after self-serve dashboard launch on April 10 and the $50K minimum drop on April 13, both since reduced to no minimum). Expect the four open questions above to either clarify or expand within the next 90 days; we track resolution on the AI Ads platform overview.
Frequently Asked Questions
#What's the difference between a context hint and a Google Ads keyword?
A Google Ads keyword is a string the matcher tries to align with words a user typed in a search query. A context hint is a 1-to-2-sentence description the matcher tries to align with the embedding of a user's active conversation in ChatGPT. Keywords match strings. Context hints match buyers. The shift from lexical to semantic matching is the single biggest mental adjustment for marketers moving between the platforms.
#How long should a context hint be?
1 to 2 sentences, under 280 characters in practice. Long enough to encode the buyer's persona, intent, and any disqualifying scope. Short enough to keep semantic precision. Hints longer than 280 characters tend to drift across multiple buyer scenarios and reduce match precision. The 280-character ceiling is not officially documented but is consistent with what the platform's UI accepts and what behavioral testing confirms.
#Can I import my Google Ads keyword list as context hints?
The dashboard does not have a keyword-list import. Each context hint goes into a free-text field per ad group, written in plain English. The keyword (2-to-5-word string) and the hint (1-to-2-sentence description) are different formats, so translation is required. The 5 Context Hint Patterns provide the canonical translation taxonomy. Our free Google Ads to ChatGPT Ads converter automates the classification, but the manual exercise builds the muscle memory.
#How many context hints should I add per ad group?
Behavioral testing suggests 1 context hint per ad group performs better than 3 or 5, though OpenAI does not publish an official recommendation. Adding more hints to a single ad group dilutes match precision because the embeddings collide and overlap. The discipline is the inverse of Google Ads' breadth-first instinct: fewer, more precise hints outperform more, less precise hints. Use multiple ad groups for multiple hints.
#Where do I enter context hints in ChatGPT Ads?
Context hints live in the self-serve ads.openai.com dashboard, which OpenAI opened to every US advertiser on May 5, 2026. The hint is a free-text targeting field at the ad-group layer. There is no keyword-list screen, no audience-list import, no demographic toggles. Context hint and bid amount are the two inputs that determine reach and price. Everything else in the campaign builder is creative and reporting.
#Do context hints work for B2C as well as B2B?
Context hints work for any vertical with a definable buyer scenario. The 5 Patterns (Persona + Intent, Question, Topic + Disqualifier, Outcome, Stack Comparison) apply to consumer purchases, household decisions, and personal goals as much as to B2B evaluations. The pattern set is buyer-scenario neutral. The only verticals where hints struggle are categories without a clear buyer: commodity products, impulse purchases, and undifferentiated repeat purchases.
#What happens if my context hint doesn't match any conversations?
The placement does not serve and the campaign accumulates zero impressions. The dashboard simply shows zero delivery for the affected ad group, with no error message. If a hint runs for several days with no impressions, the most likely causes are: a hint too narrow to match active conversations, a topic outside categories ChatGPT users actively discuss, or a scope that excludes too much. Tightening persona and broadening scope usually resolves it.
Related Reading
- How to Convert Google Ads to ChatGPT Ads: A 2026 Migration Methodology
- How to Convert Google Ads to ChatGPT Ads in the AI-Advisors Platform
- Inside the ChatGPT Ads Dashboard: A First Look at CPC Setup, Bidding, and Context Hints
- How Do ChatGPT Ads Work? Format, Context Hints, and the AI Visibility Lift (May 2026 Update)
