The short answer
For B2B advertisers deciding whether ChatGPT Ads can be optimised without a search terms report.
- 1ChatGPT Ads matches your offer against a live conversation, then reports impressions, clicks, CTR, CPC, spend and conversions. It never shows the conversation.
- 2There is no search terms report, no prompt themes, no negative keywords. You cannot see or exclude what triggered an impression.
- 3That hurts B2B more than B2C: small buying committees, long cycles and low volumes mean wasted delivery is invisible for months.
- 4You still have seven controls: context hints, creative, landing page, geography, bidding, budget pacing and the conversion event you optimise to.
- 5So optimise by hypothesis. One change per ad group, held long enough to read, judged on CRM outcomes rather than platform conversions.
The argument
Every performance channel is a loop. You form a hypothesis about who should see the ad, you spend, you look at what the machine actually did, and you correct. Google Search gave us that loop in its purest form: you bid on a keyword, and a search terms report told you the literal words people typed. When the report said you were paying for "free" and "jobs", you added negatives and the loop closed.
ChatGPT Ads gives you the first half and withholds the second. You write context hints describing the conversations where your offer belongs. The system serves. You get impressions, clicks, spend, click-through rate, cost per click and conversions. What you never get is any view of the conversations, prompts or themes that produced those impressions.
The matching may well be sophisticated. It's the advertiser's controls and feedback loop that are primitive.
Most of the criticism I read gets this the wrong way round. People say the targeting is crude. I don't buy that. My read is that the targeting is doing something fairly clever and simply won't tell us what.
What the interface actually shows you
Here's the inventory from the accounts we're running right now. Left column exists. Right column doesn't, and probably never will.
- Impressions, clicks, spend
- Click-through rate, average CPC, average CPM
- Conversions, and ROAS on supported order events
- Performance by ad
- Broad device split, mobile and desktop
- Delivery and review status
- The conversation the ad appeared in
- The prompt or question that triggered it
- Any themes or clusters behind delivery
- A search terms equivalent of any kind
- Which part of your context hint did the work
- Anything resembling a negative keyword control
Don't sit around waiting for the right hand column to get filled in. Those conversations are private, and the whole product depends on people believing that. Plan for the constraint sticking around in some form.
The one report that doesn't exist
A search terms report was never really reporting. It was the diagnosis. It's the one thing that tells you whether a bad month was a bad audience, a bad message or a bad offer. Take it away and all three failures look exactly the same from the dashboard: money out, nothing back.
Search Engine Land compared the experience to the early black box era of Performance Max, and that comparison is fair with one important difference. Performance Max at least eventually surfaced asset group and placement signals. Here the missing information is the user's private conversation, which is the one thing the platform is structurally unlikely to hand over.
Why this bites B2B harder
A consumer advertiser with a fifty pound product gets a verdict in a week. Volume covers for the missing diagnostics: enough conversions arrive that the numbers themselves become the signal. B2B has none of that cover.
Five leads a month is a normal, healthy result for a niche B2B advertiser and a statistically useless dataset. The absence of qualitative signal isn't offset by quantity.
The thing you actually want to know, whether the lead became pipeline, arrives months after the spend decision. You're flying on proxies for a quarter.
Ads serve to Free and Go users. Plus, Pro and Business accounts never see them. In some markets that quietly removes the buyer, and nothing in the interface will tell you that's what happened.
At three to eight pounds a click, a month of undiagnosable delivery costs real money. Search Engine Land's own tests observed CPCs around two to five dollars.
A person asking how to fix a problem and a person asking who to hire to fix it look similar to a semantic matcher and are worth wildly different amounts to you.
This is also why the published results are all over the place. One agency reports 150 plus qualified leads across three B2B SaaS clients at a cost per lead around 60% under Google. Another reports roughly $415 spent, 8,940 impressions, 60 clicks, not one interested buyer. I'd believe both of them. Neither can tell you why, because neither was ever shown the demand they paid for.
Want a straight read on whether this channel suits your business?
Book a 30 minute call. We'll look at your organic ChatGPT referrals and your buyer's likely plan tier before anyone spends anything.
The seven levers you do have
Accepting the constraint doesn't mean rolling over. It changes what counts as a lever. Here's the whole toolkit, and every item on it is a guess you're testing rather than a dial you're setting.
- Campaign structure. One objective, one geography, one offer. Anything blended is unreadable.
- Ad group separation. One conversation state per group. This is your only means of isolating a hypothesis.
- Context hints. The description of the situation. Concise beats exhaustive, and over-engineering strangles delivery.
- Creative. The single strongest filter you control. A specific ad self-selects the right person even from broad delivery.
- Landing page. Part of the relevance model, not just the destination. Specific page, specific intent.
- Bid strategy and budget. How aggressively you buy the hypothesis while you test it.
- First-party audiences. Uploaded lists to include known prospects, exclude existing customers, or apply bid multipliers.
Look at what's actually on that list. It's nearly all message and structure. Barely any of it is targeting in the way a LinkedIn buyer would understand the word. So the people who get good at this channel will be writers, not audience builders.
How to optimise without a search terms report
So you build your own diagnostics downstream. You'll never see the conversation, but you can get a detailed enough picture of who turned up to work backwards from it.
- Run structure as the experiment. Two to four ad groups, each a genuinely distinct conversation state, all else held constant. The group that delivers and converts tells you which intent exists.
- Instrument engagement, not just conversion. Engaged session rate and average engagement time move weeks before conversion data is thick enough to read.
- Ask the human. A "how did you hear about us" field and a "what were you trying to solve" question on the form. On a channel with no query data, the lead's own words are the closest thing to a search terms report you'll ever get.
- Log the qualitative. Every sales call from this channel gets a note on what the person was actually researching. Twenty of those notes beat any dashboard you could build.
- Change one variable at a time. Painful and slow. Also the only method that produces knowledge rather than opinion.
That last one is where the black box actually hurts. You can still optimise. But every lesson costs you a full test cycle and a few hundred quid, when Google would have handed you the same lesson in a report on a Tuesday afternoon.
The fair case for the black box
I'm not accusing OpenAI of anything cynical here. There's a decent argument on the other side and it deserves a fair hearing.
- These are private conversations. People type things into ChatGPT they would never type into Google. Handing advertisers a themed report on that content would be a serious breach of the trust the product runs on.
- Keyword reports trained a generation of advertisers to optimise the wrong unit. Plenty of Google accounts were destroyed by people negating their way to a tiny, safe, irrelevant audience.
- The platform is young. Conversion measurement, custom audiences, automated bidding and conversion bidding have all arrived quickly, and it hit a billion dollar annualised run rate inside 200 days. This isn't a hobby product.
All fair. None of it changes where an advertiser stands this month, which is paying for a decision they're not allowed to look at. My read: test it, test it properly, and stop pretending you can run it like a search account.
Interesting enough to test. Not mature enough to trust. Both halves of that sentence matter.
We publish what we find, including the tests that go nowhere. In 2026 that's about the only useful thing anyone can add to this channel.
Read next
Common questions
We run ChatGPT Ads and publish what happens
- A suitability read before spend: organic ChatGPT referrals, buyer plan tier, click price tolerance.
- Account, Pixel, CAPI and CRM source capture built so the downstream data replaces the missing diagnostics.
- A structured hypothesis test across two to four conversation states, one variable at a time.
- Qualitative capture on every lead, so you learn what people were actually researching.
- A thirty day scale, extend or stop call measured against your existing cost per qualified lead.
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. This channel is young and we don't think anyone should pay a premium while the whole industry is still learning it.