Context Hints Are Not Keywords. Here Is the Translation Table
By Ansh Khandelwal 6 min read
In brief
Translate Google Ads keywords into ChatGPT context hints and account for the loss of match types, negatives, search terms, and quality score.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 October 2026
If you run Google Ads, you already know how to think about intent. What you do not have is a place to put half of what you know.
ChatGPT Ads has no keywords. It has context hints, which are ad-group-level descriptions of the situations where your product is useful, matched against the meaning of a live conversation rather than the text of a query. OpenAI's Create Ad Groups documentation, updated August 2026, frames them as descriptions of "what your product offers, who it helps, or when it may be useful".
Most of your account concepts survive the move. Four do not, and those four are where first campaigns quietly fail.
The translation table
| Google Ads | ChatGPT Ads | What changes |
|---|---|---|
| Campaign | Campaign | Same job. Objective, budget, dates, locations |
| Ad group | Ad group | Same job, higher stakes. It is the smallest unit that gets reported |
| Keyword | Context hint | Moves from string matching to meaning matching. Set per ad group, not per keyword |
| Match type | Nothing | No broad, phrase or exact. Breadth is controlled by how you write the hint |
| Negative keyword | Nothing | No exclusion layer at all |
| Search terms report | Nothing | You never see what conversation you appeared in |
| Quality score | Nothing visible | Relevance is weighted in the auction but never surfaced as a number |
| Max CPC bid | Max bid, ad group level | Familiar. OpenAI guidance has circulated around three to five dollars for CPC, reported rather than published as a floor |
| Location targeting | Location targeting | Country, region, DMA, postal code. Closely comparable |
| Audience lists | Custom audiences | First-party upload only. No third-party segments, no behavioural retargeting |
| Responsive search ad | Ad | One headline, one description, one image, one URL. Far tighter limits |
| Conversion tracking | OpenAI pixel and Conversions API | Familiar shape, much younger implementation |
| Auction | Relevance-weighted second price | Same family. Relevance carries more weight and is less visible |
The four blanks
No match types. On Google, breadth is a setting. Here it is a writing decision, made once, invisibly, when you choose how much of a situation to describe. A hint reading "project management software" is a broad match you cannot dial back. A hint carrying an audience, a trigger and a constraint is close to phrase match. Nothing in the interface tells you which you have written.
No negatives. There is no exclusion layer. You cannot block a theme, a competitor comparison, or the DIY version of your own question. What replaces it is structural: separate low-fit themes into their own ad groups so they can be paused, and let the creative do the filtering by naming a price, a company size or a qualifying condition that the wrong reader will not click.
No search terms report. This is the one that changes daily practice most. Google account management is largely a loop of reading search terms, adding negatives and reallocating. That loop does not exist here. There is no query mining, no waste list, no discovery of the profitable phrasing you had not thought of. Optimisation moves to the only levers that report: ad group structure, creative variants and landing page.
No quality score. Relevance is weighted in the auction, so a better-matched ad and page can serve more and cost less, but no number tells you where you stand. You infer it from CPC moving against a fixed bid.
What a pasted keyword list actually does
Here is the part that contradicts the tidy story.
In 2026 Search Engine Land ran two ad groups with identical creative, one filled with a Google-style keyword list and one with full natural-language questions. The keyword group served far more, with a slightly better CTR and a slightly lower average CPC.
One test, one advertiser, one moment, reported by a third party and not replicated by us. But it means a pasted keyword list is not inert, and the confident claim that keywords "do not work" on this platform is not supported by the only public head-to-head there is.
The reading that fits both the documentation and the test: the matcher works on meaning, and a dense keyword list does carry meaning, just without constraints. It buys breadth. On a channel where you cannot see what you matched and cannot exclude anything, breadth is the expensive option, which is an argument for written hints on economics rather than on the platform rejecting keywords.
What we cannot tell you
- How the matcher weights a hint against a conversation. Undocumented.
- Whether match breadth can be inferred from hint length. Not measurable without hint-level reporting.
- What relevance weighting is worth in cost terms. The mechanism is documented, the magnitude is not.
- Whether the Search Engine Land result generalises. One test, one category.
Quick answers
Does ChatGPT Ads use keywords? No. It uses context hints, set at ad group level, describing situations rather than search strings.
Can I paste my Google keyword list into a context hint field? You can, and one published test found it served more and cost slightly less than written hints. It buys breadth without constraints, which is expensive on a channel with no exclusion layer.
Are there match types in ChatGPT Ads? No. Breadth is decided by how much of the situation your hint describes.
Can I add negative keywords to ChatGPT Ads? No. Separate low-fit themes into their own ad groups and use creative to disqualify the wrong reader.
Is there a search terms report? No. You cannot see which conversations you appeared in, which removes the core Google optimisation loop.
Does ChatGPT Ads have a quality score? Not one you can see. Relevance is weighted in the auction but never shown as a number.
Sources
| Claim | Source | Tier |
|---|---|---|
| Context hints are ad-group-level natural-language descriptions | OpenAI Help Center, Create Ad Groups for ChatGPT Ads, August 2026 | Confirmed, primary |
| Campaign, ad group, ad structure with objective, budget, dates and locations at campaign level | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Location targeting supports country, region, DMA and postal code | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Custom audiences are first-party upload | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| No match types, negatives, search terms report or visible quality score | OpenAI documentation, by absence | Absent |
| Keyword-style ad group outserved and outperformed a conversational one | Search Engine Land, 2026 | Reported, third party, single test |
| Three to five dollar CPC starting bid guidance | Multiple agency write-ups citing OpenAI guidance, 2026 | Reported |
Related reading
- How to Write a ChatGPT Ads Context Hint
- ChatGPT Ads Context Hints: What They Are, and What They Are Not
- How the ChatGPT Ads Auction Works
- ChatGPT Ads vs Google Ads vs Meta Ads
Changelog
12 September 2026, v1.0. First publication. Introduces the four blanks framing and records the Search Engine Land keyword-list result against the category's standard claim.
Field kit
Tools
Site
OpenAI (primary)
Ansh works across GEO strategy, B2B research, and execution. At InPromptAds, he translates new AI advertising products into clear operating advice, tests, and measurement questions for marketing teams.
