Mapping Your Google Ads Keyword Clusters to Context Hints
By Ansh Khandelwal 8 min read
In brief
One Google ad group becomes one ChatGPT context hint theme, and the raw material is your top converting search terms, not your keyword list.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 October 2026
The conversion is one to one at the ad group level and many to one below it. A Google ad group becomes a ChatGPT ad group. The twenty or two hundred keywords inside it collapse into a small set of context hints describing the situation those keywords were proxies for.
This is the procedure. The conceptual comparison lives in our translation table piece and is not repeated here.
Before you start: pull three exports, not one
Open the Google Ads account and export these, dated, for the last 90 days:
- Ad group performance with conversions and cost per conversion.
- Search terms report, filtered to terms with at least one conversion.
- Ad copy for the two highest-converting ads in each ad group.
The second export is the one that matters. Keywords tell you what you bid on. Converting search terms tell you what a real person actually typed on the way to becoming a customer, which is closer to what someone says to ChatGPT. If your account runs mostly broad match with Smart Bidding, the keyword list may be almost content-free and the search terms report is the only usable input.
Step 1: rank ad groups and cut to five
Sort ad groups by conversions descending. Delete from consideration anything that converted fewer than ten times in 90 days, anything branded, and anything whose entire value comes from a competitor's name.
Branded groups have no counterpart on ChatGPT Ads. A hint saying "people looking for [your brand]" describes no situation, and there is no exclusivity mechanism to protect the term. Competitor groups sit inside an undocumented policy area we treat separately.
You should be left with three to five ad groups. That is the whole first build.
Step 2: write the situation sentence
For each surviving ad group, look only at its converting search terms and answer four questions in writing. Do not look at the keyword list while you do this.
- Who is the person? Role, company size, sector.
- What has just happened to make them search today? A renewal, a failed audit, a hire, a complaint, a deadline.
- What are they trying to decide, phrased as they would phrase it, not as you would.
- What disqualifies them? A geography, a budget floor, a platform they are not on.
Write one sentence containing all four. That sentence is a candidate hint. It will feel too long. Length is not the problem, vagueness is.
Step 3: the worked example
A real-shaped B2B example, invented for illustration and carrying no account data.
| Input | Content |
|---|---|
| Google ad group | Expense management software, exact and phrase |
| Keywords | expense management software, expense report software, corporate expense tool, spend management platform |
| Converting search terms | expense software that syncs with netsuite, replace concur alternative, expense reports taking finance team too long, expense tool for 150 employees |
| Who | Finance lead or controller, 100 to 500 employees |
| What just happened | Month-end close is slipping and finance is doing manual receipt chasing |
| Deciding | Whether to replace an incumbent tool or add automation to the current one |
| Disqualifier | Runs on NetSuite, not a startup on spreadsheets |
| Hint | A finance lead at a 100 to 500 person company whose month-end close is slipping because expense reports are chased manually, deciding whether to replace an incumbent expense tool that does not sync cleanly with NetSuite |
Notice what did not survive: the word "software" as a standalone product category, and the phrase "spend management platform". Those were bidding instruments. The situation carries them implicitly.
Notice what did survive from the search terms and not from the keywords: NetSuite, the 150-employee scale, and the incumbent-replacement framing. All three came from the search terms export. None appeared in the keyword list.
Step 4: split when the situations diverge, not when the keywords do
Google rewards granular ad groups. ChatGPT Ads does not, because the ad group is the smallest reporting unit and splitting divides already thin data.
Split only when the answer to "what has just happened" is genuinely different. In the example above, "replacing an incumbent tool" and "buying a first expense tool" are two situations, two buying committees and two pieces of creative. They split. "Expense report software" and "expense management software" are one situation. They do not.
A useful ceiling for a first build: three to five ad groups, two to four hints each.
Step 5: check the hint against four failure tests
Read each hint and reject it if any of these is true.
- The product category test. Remove your product name and category from the hint. If nothing meaningful remains, you have written a keyword in a longer form.
- The competitor test. Could a direct competitor use this hint unchanged? If yes, it describes a market, not your position in it.
- The verifiability test. Does it contain at least one checkable detail: a system, a headcount band, a regulation, a job title, a trigger event?
- The conversation test. Read it aloud as something a person would actually be discussing with an assistant. "Enterprise-grade scalable solutions" fails. "Our close is taking eleven days and finance is burning out" passes.
Step 6: carry the disqualifier into the ad, not just the hint
There is no negative keyword layer, so the disqualifier you wrote in step 2 has a second job. Put it in the ad copy as a checkable condition: a price floor, a headcount band, an integration requirement. On this channel the creative is the only exclusion control you have.
The honest problem with this procedure
Every step above assumes converting search terms are a good proxy for conversational context. That assumption is untested in public.
There is one piece of published evidence, and it cuts the other way. Search Engine Land ran a test in 2026 with two ad groups, identical creative, one filled with Google-style keywords and one with full natural-language questions. The keyword group served far more, with slightly higher CTR and slightly lower average CPC. One test, one advertiser, reported by a third party.
If that result generalises, the elaborate translation above may buy precision at the cost of volume. The defensible reading is that written hints trade reach for fit, and on a channel with no exclusion layer and no search terms report, fit is the thing worth paying for. That is a judgement, not a measurement, and we label it as ours.
What we cannot tell you
- Whether a translated hint outperforms a pasted keyword list in your category. No per-hint reporting exists, so you cannot A/B two hints inside one ad group and read the result.
- How much of a hint the matcher actually uses. Undocumented. Nothing published says whether a 40-word hint is read in full or reduced.
- Whether converting search terms predict conversational context. Not measurable. There is no search terms report on ChatGPT Ads to compare against.
- How many hints is too many. OpenAI publishes no cap or guidance on hint count per ad group.
Quick answers
How many Google ad groups should I migrate first? Three to five, chosen by conversion volume in the last 90 days. Drop branded and competitor groups entirely. A larger first build divides data across ad groups that are already the smallest reporting unit.
Do I use my keyword list or my search terms report? The search terms report, filtered to terms with at least one conversion. Keywords record what you bid on. Converting search terms record what a buyer actually typed, which is closer to conversational phrasing.
How many context hints per ad group? Two to four for a first build. Each should describe a different facet of the same situation, not a different situation. A different situation is a different ad group.
Should I keep my SKAG structure? No. Single keyword ad groups have no counterpart here. The ad group is the smallest unit that reports, so granularity that made sense on Google splits your data into unreadable fragments.
What do I do with branded keywords? Nothing. There is no brand campaign equivalent, no exclusivity, and a hint describing people searching your brand name describes no situation the matcher can use.
Can I just paste the keyword list in? You can, and one published 2026 test found it served more and cost slightly less. It buys breadth without constraints, which is expensive where you cannot see or exclude what you matched.
Sources
| Claim | Source | Tier |
|---|---|---|
| Context hints are ad-group-level natural-language descriptions of when a product is useful | OpenAI Help Center, Create Ad Groups for ChatGPT Ads, August 2026 | Confirmed, primary |
| Ad group is the smallest unit with bid, default destination URL and context hints | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Reporting exists at campaign, ad group and ad level, not at hint level | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| No keywords, match types, negatives or search terms report | OpenAI documentation, by absence | Absent |
| Keyword-style ad group outserved and outperformed a conversational one | Search Engine Land, 2026 | Reported, third party, single test |
| Converting search terms are the best available proxy for conversational context | InPromptAds | Inference, ours |
Related reading
- Context Hints Are Not Keywords: The Translation Table
- How to Write a ChatGPT Ads Context Hint
- Can You Exclude Conversations on ChatGPT Ads?
- Context hints generator
- How Many Ad Groups Should a First ChatGPT Campaign Have?
Changelog
12 September 2026, v1.0. First publication. Establishes the six-step keyword-cluster to hint-theme procedure and the four failure tests, with the Search Engine Land result recorded against it.
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.
