The short answer
Context hints are ad group level descriptions of the conversations and situations where your offer is a sensible next step. OpenAI is blunt about what they aren't: not exact match keywords, not hard audience rules, no guarantee your ad turns up next to particular wording. They're a hypothesis, read alongside your creative and your landing page.
In practice: one theme per ad group, short rather than exhaustive, and tested rather than assumed. A few hundred words on business type, intent and use case is the shape that works. One advertiser wrote roughly a thousand words of audience manifesto and watched delivery dry up almost completely.
Why they aren't keywords
A keyword is a contract. You bid on it, a match type decides how loosely it applies, a search terms report shows you what it caught, and a negative lets you carve out the rubbish. Four mechanisms, all of them binding on the platform.
A context hint gives you none of that. No match type, no report, no negatives. You're briefing a very fast, very literal media buyer who will never once tell you which part of the brief they actually acted on.
A keyword is a rule the platform must obey. A context hint is a description the platform may agree with.
So the keyword habits travel badly. Volume data is beside the point. Match type strategy doesn't exist. And long tail expansion, the thing we've all spent a decade getting good at, actively works against you here: more terms in the box makes the hypothesis muddier, not more complete.
What the public tests found
Two data points get quoted everywhere and they flatly disagree. That's the reason this page exists.
A controlled test found a simpler list of keyword-like phrases delivered more often and marginally outperformed a fully conversational, question-led hint setup.
An approximately 1,000 word hint setup barely delivered at all. Cutting back to a few hundred words focused on business type, intent and use cases increased delivery.
The wrong lesson is "always use keywords" or "always use prompts". Look at what the two results share instead: the shorter version won both times. That's one finding and two open questions. Not enough to build a method on.
The three hint styles
Three candidates worth putting head to head. Same offer, same creative, same landing page, three ad groups, one difference between them.
LinkedIn Ads, LinkedIn advertising, B2B lead generation,
LinkedIn ads agency, LinkedIn campaign management,
B2B paid social, cost per lead LinkedIn, LinkedIn ads not workingFamiliar, compact, easy to write. The Search Engine Land test favoured this shape. The risk is that it describes a subject rather than a situation, so it may match people reading about the topic rather than living the problem.
Why aren't my LinkedIn Ads generating leads?
Is my LinkedIn targeting wrong?
Should I hire a LinkedIn Ads agency or do it in-house?
How much should a B2B company spend on LinkedIn Ads?
Why is my cost per lead so high on LinkedIn?Mirrors how people actually talk to the model. Intuitively the best fit, and the one that underperformed in the test everyone quotes. Worth understanding why before dismissing it: a question set may be too narrow, describing five conversations rather than a class of them.
B2B marketing leaders currently running LinkedIn Ads who are
frustrated by poor lead volume, high cost per lead, weak creative or
campaigns producing clicks without qualified pipeline. Relevant when
users are working out why their LinkedIn campaigns are underperforming,
whether their targeting or offer is wrong, or whether they need
specialist help improving existing campaigns.Describes who the person is, what's going wrong and where they're in the decision. This is the shape OpenAI's own guidance points toward, and the one we default to on new accounts. It's also the hardest to write well.
Want your context hints written and tested properly?
Book a 30 minute call. We'll show you the structure we use and what it has produced so far.
The test we're running
We're running this on our own account first. Charging a client to learn something feels wrong. Here's the design, so you can copy it or pick holes in it:
| Element | How it's set |
|---|---|
| Structure | One campaign, three ad groups. Only the hint style changes between them. |
| Creative | Identical set of three ads in every group. Problem, specialist, proof. |
| Landing page | The same specific page for all three, never the homepage. |
| Bidding | CPC, equal budgets, no conversion bidding while the sample is thin. |
| Duration | Thirty days, no mid-flight edits however tempting. |
| Primary measure | Qualified leads and what those leads say they were researching. |
| Secondary | Delivery share, click-through rate, engaged session rate, cost per click. |
One deliberate choice there. Delivery volume is a secondary measure, never the verdict. A hint style that delivers twice as much and converts nobody has lost. With no search terms report to check against, delivery is the number most likely to walk you into the wrong conclusion.
We'll publish the numbers on this page when the thirty days are up, including if the answer is that all three performed the same. A null result is still information, and on a channel this young it's more useful than another confident blog post.
How to write one that delivers
Whichever style you land on, the same rules apply. A good hint says who the person is, what's going wrong for them, how far into the decision they've got, and the use cases that matter. Nothing else goes in.
- Who and what context
- The specific problem
- Stage of the decision
- Two or three real use cases
- Optional example themes
- Thousand word manifestos
- Age and job title stuffing
- Services this ad group isn't selling
- Every pain point you can think of
- Anything you wouldn't say to a new hire
The most useful habit I've found: write it, read it back, and ask whether it describes one person in one moment. If two quite different people fit the description, split it into two ad groups. That's most of the craft, and it sits much closer to positioning work than keyword research.
Structure does the work keywords used to
Here's where I think most people land eventually. You can't see what a hint caught, so the only way to learn anything is to make the container narrow enough that the result can only mean one thing. Ad group structure is now the diagnostic instrument.
- Existing solution failing. They have something in place and it isn't working.
- Comparing vendors. Actively evaluating, wants a basis for the decision.
- A specific technical problem. Deep in one issue, not yet thinking about suppliers.
- Looking for a specialist. Already decided to get help, working out who.
Four ad groups, four hypotheses, four results you can actually read. Shove all four descriptions into one group called "B2B decision makers" and whatever happens next, you'll never know why it happened. Which lands where every part of this channel lands: the discipline sits in how tightly you frame the question, not in the targeting box.
Context hints may or may not become the new keyword research. Ad group structure has already become the new search terms report.
Read next
Common questions
Want your context hints built as a real test rather than a guess?
- Ad group structure built as distinct conversation states, so every result is interpretable.
- Three hint styles written and tested against each other with everything else held constant.
- Creative built to self-select the right person, since there's no negative keyword control.
- Qualitative capture on every lead, so you learn what people were actually researching.
- A written readout of what won, what lost, and what it means for your next month of spend.
Starter £1,250/mo, Growth £1,725/mo, Scale £4,025+/mo. For current clients we cover setup and management of the first test and you fund the media. Everything we learn goes into a benchmark set that improves the next account we build.