The ChatGPT Click Arrives Mid-Conversation. Most Landing Pages Restart It
By Ansh Khandelwal 8 min read
In brief
Design landing pages for ChatGPT Ads visitors who arrive mid-research, with continuity instead of a category introduction.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 December 2026
Build the page to resume, not to restart. Someone arriving from a ChatGPT ad has spent several turns articulating their problem in full sentences before your ad appeared under the answer. A page that opens by explaining what category you are in is asking them to do that work again.
This is the structural difference between this channel and search, and almost nothing in standard landing page practice accounts for it.
Why a ChatGPT click is not a search click
A search click begins with two to four words typed into a box. Everything else about that person's situation is inferred. A ChatGPT click begins with a conversation: a described problem, usually a constraint, often a rejected alternative, and a model response that has already narrowed the field before your ad appears below it.
OpenAI's own ad selection documentation says the system considers "the context and intent of your current conversation" alongside the ad's landing page, title and copy. The conversation is the input. The click is the output of a process that has already done qualification work you did not pay for and cannot see.
What you actually know about the arrival
You know four things with reasonable confidence, none of them from reporting.
You know the person described a situation rather than named a product, because that is what the conversational interface invites. You know a model gave them an answer immediately before your ad, so they arrive post-explanation rather than pre-explanation. You know your context hint matched the meaning of that conversation, so the ad group's theme is a real description of their state. And you know they clicked past an answer they already had, which means the answer was incomplete in some specific way.
What you do not know is the wording. There is no search terms report, no conversation data, no adjacency reporting. You know the theme. You never know the sentence.
That asymmetry is the whole design problem: high confidence about the topic, zero visibility into the phrasing.
The reset page, and why it is the default
Call the failure the reset page. It opens with a category definition, a value proposition written for a stranger, and a nurture path that assumes the visitor has just learned your category exists.
It is the default for a boring reason. Most B2B landing pages were written for cold paid social or for a homepage-adjacent audience, where the assumption of zero context is correct. Pointing a ChatGPT ad group at one of those pages is not a mistake anyone notices, because the page looks fine. It just spends the first screen re-establishing something the visitor established for themselves three minutes ago.
OpenAI's Create Ads guidance is explicit that advertisers should "link to the most relevant destination: use product pages, collections, or content pages rather than defaulting to a generic homepage". That is a statement about page selection. The reset page problem survives correct page selection, because a dedicated page can still open in a register calibrated for someone who knows nothing.
The structural response: what the first screen has to carry
Three things, in this order, above the fold.
A resumption line. One sentence that names the situation the ad group's hint theme describes, in the second person, as a condition already true. Not "Expense management for growing teams" but "You are chasing receipts over Slack in the last week of every month and it has started failing audit samples." The visitor should recognise their own last few turns in it. This is the cheapest signal available that the page is a continuation rather than a fresh start.
The specific claim. Not the category benefit. The thing the model's answer could not give them: a price, a mechanism, a constraint you satisfy that the generic answer did not address. The model already told them expense tools exist. It did not tell them which one reads their existing approval chain without a migration.
One action calibrated to depth. A visitor who has just been given an answer by a model is in evaluation, not in purchase and not in discovery. The action should be the next honest step from evaluation, which is usually something they can do without a sales conversation. Article 31 in this series covers the ladder.
Everything the reset page does in the first screen, the category framing and the trust logos and the three-column feature grid, moves below that.
Depth is a band, not a point
The trap in all of this is over-fitting. You know the theme, not the turn. Two people can match the same hint from very different depths: one has been comparing vendors for a week, one asked a general question ten minutes ago.
So write the resumption line to the deepest plausible arrival and make the recovery cheap for anyone shallower. In practice that means the situation sentence sits at the top and a single plain line underneath it says what the product is, for the reader who has arrived earlier in the process than you assumed. That line costs you nine words. Omitting it costs you every visitor whose conversation was broader than your hint.
The reckoning: warmer traffic that converts worse
The argument above says these clicks arrive warm. One agency's published numbers say the paid ones do not perform like it.
Orbit Media's 2026 study of GA4 data from 97 B2B and lead generation sites, covering 1 July 2025 to 30 June 2026 and 28.9 million sessions, found ChatGPT referral traffic converting at 2.08% on high intent actions against 0.5% for Google Search. That is third-party, dated, and not a vendor claiming credit for its own service. It supports the warmth thesis.
But that is organic referral traffic, not ads. Q1Media, reporting on its own live campaigns in July 2026, put ChatGPT ads conversion at roughly 2%, which it described as about half the rate of organic ChatGPT traffic, and attributed the gap to lower intent in a placement the user did not choose. That is a vendor account of its own work and it has not been replicated.
Held together, the defensible reading is narrower than the headline: the conversation before the click is real context, but an ad appearing below an answer is an interruption of that conversation, not a recommendation inside it. The visitor is topically warm and commercially uncommitted. A page that assumes warmth and asks for a sales call is misreading the second half of that sentence.
What we cannot tell you
- What the visitor actually typed. No search terms report, no conversation data, no adjacency reporting. Absent by design.
- How deep into a conversation the click happened. Turn count is not reported and is not inferable from anything in Ads Manager.
- Whether resumption-style pages outperform reset pages on this channel. No published controlled test exists, and InPromptAds has no first-party campaign data.
- Whether the Q1Media paid-versus-organic gap generalises. One agency, its own accounts, one point in 2026.
- How much of the landing page OpenAI's selection model actually reads. OpenAI names the landing page as a signal. It does not say what it extracts from it.
Quick answers
Is a ChatGPT ad click warmer than a search click? Topically, yes: the user described their situation in sentences before the ad appeared. Commercially, less so. The ad interrupts an answer rather than being chosen from a results page, and one agency reports paid converting at about half its organic rate.
Should I send ChatGPT ads traffic to a dedicated page? Yes, and OpenAI's own guidance says to link to the most relevant destination rather than a generic homepage. Dedicated is necessary but not sufficient: a dedicated page can still open as though the visitor knows nothing.
What goes above the fold? A resumption line naming the situation your hint theme describes, the specific claim the model's answer did not give them, and one action appropriate to evaluation stage.
What if my hint matches people at different stages? Write to the deepest plausible arrival and add one plain line saying what the product is, for anyone who arrived earlier than you assumed.
Can I personalise the page to the conversation? No. Nothing about the conversation is passed to your site. The only variable you control is which ad group points at which page.
Does the landing page affect whether my ad serves? OpenAI lists the ad's landing page among the signals used to select ads. It publishes no score, so the direction is documented and the magnitude is not.
Sources
| Claim | Source | Tier |
|---|---|---|
| Ad selection considers conversation context and intent, and the ad's landing page, title and copy | OpenAI Help Center, Ads in ChatGPT, 2026 | Confirmed, primary |
| Link to the most relevant destination rather than a generic homepage | OpenAI Help Center, Create Ads for ChatGPT, 2026 | Confirmed, primary |
| No search terms report, conversation data or adjacency reporting | OpenAI Ads Manager documentation, by absence | Absent |
| ChatGPT referral traffic converted at 2.08% on high intent actions versus 0.5% for Google Search, 97 B2B sites, 28.9m sessions, July 2025 to June 2026 | Orbit Media Studios, AI traffic conversion research, 2026 | Reported, third party |
| ChatGPT ads converting around 2%, roughly half the rate of organic ChatGPT traffic | Q1Media, live campaign write-up, July 2026 | Vendor |
| Dedicated pages convert at 5 to 15% against homepages at 1 to 3% | Lapis, ChatGPT ads landing page guide, updated August 2026 | Vendor |
| The resumption line and the depth band correction | InPromptAds method | Inference, ours |
Related reading
- Homepage or Dedicated Page for ChatGPT Ads Traffic
- Continuity: Making the Page Answer the Question the Ad Interrupted
- Forms for Research-Stage Clicks
- How to Write a ChatGPT Ads Context Hint
- The Psychology of Conversational Ads
Changelog
12 September 2026, v1.0. First publication. Establishes the reset page as the default failure and the resumption line as the structural fix, with the Orbit Media and Q1Media figures set against each other rather than merged.
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.
