Updated 25 October 2026.
In ChatGPT Ads, the landing page comes into play long before the click: OpenAI documents it among the signals that decide which ad appears under an answer, on the same footing as context hints, title and body. A page that does not carry the conversation forward costs you twice, because it is selected less often and converts less when it is.
Why does the page count in selection?
Because the system reads the destination, not just the ad. OpenAI lists the conversation context, the ad group’s context hints, the title, the body, the landing page and overall relevance among its selection signals, all inside a second-price auction weighted by relevance.
In practice, a page aligned with the intent raises relevance, and higher relevance can win the auction with a lower bid, so the landing page works on your buying cost before it works on conversion.
There is also a readability condition. The page must be reachable by OpenAI’s crawlers, and a page behind a login, blocked in robots.txt or rendered only by client-side JavaScript cannot be read, so it cannot be assessed.
The homepage and its clicks without conversions
The homepage underperforms because it answers everyone, which means it answers no one. The user has just asked ChatGPT a precise question, and what they get back is a general positioning, a menu and six promises, none of which picks up the thread they started in the chat.
| Homepage | Page per intent | |
|---|---|---|
| H1 | The brand and its mission | The user’s question, rephrased |
| Above the fold | Positioning | Answer to the intent, proof, CTA |
| Navigation | Full | Reduced or absent |
| CTA | Several | One |
| How the system reads it | Diffuse signal | Signal aligned with the hints |
| Typical diagnosis | Clicks without conversions | Conversion readable by intent |
The native report gives only six columns: impressions, clicks, spend, CTR, average CPC, average CPM. If every ad leads to the homepage, you will never learn which intent converts; with one page per intent, each URL becomes a readable line in your analytics tool.
One page per family of intent, really?
Yes. One ad group maps to one family of intent, and one family of intent to one page, which rules out both a page per ad and a single page for the whole account.
The families are always the same: problem, comparison, alternative, purchase, setup, price, urgency, local. Each expects a different page. Someone comparing wants a table, someone with a problem wants a use case, and someone asking about price wants the price. Simple.
- List your ad groups and the family of intent for each, then assign every family an existing URL or one to create.
- Write each page’s H1 from the ad group’s main context hint.
- Add UTMs to every ad link, with
utm_termequal to the family of intent. - Check that every page passes the checklist in the next section.
If two ad groups lead to the same page, ask yourself whether they are really two intents, because if they are not, you probably have one ad group too many.
What the page must contain
Eleven points, to check before you submit the campaign for review. They follow the method we apply on our own accounts.
| Point | What we check | Why |
|---|---|---|
| Above the fold aligned | The H1 restates the ad’s intent | Continuity of the conversation |
| Precise promise | An outcome, not a mission | Qualifies and avoids the disappointed click |
| Simple explanation | What it is, for whom, in a few lines | The user comes from a question, not from an ad |
| Fast proof | A dated figure, a customer, a review, a label | Trust before the scroll |
| Comparison | A table if the intent is comparison | It is what the user asked for |
| Use case | An example if the intent is a problem | Shows the solution applied |
| Price or next step | Price shown, or the next step obvious | Price intent does not forgive absence |
| Single CTA | One action, or a clear hierarchy | One choice can be measured, five cannot |
| Clean UTMs | utm_source=chatgpt, all five parameters |
Without them, the click is invisible |
| Pixel or Conversions API tested | A real click fires the event | Without them, the conversion is invisible |
| Reachable by OpenAI’s crawlers | No blocking, no login, content in the HTML | Otherwise the page does not enter selection |
The first eight points are the craft of the page itself. The last three are operating conditions: without them the click stays invisible, so does the conversion, and the page may never be assessed at all. Tracking is covered in measuring ChatGPT Ads conversions.
From prompt to page: three pairs
A good page answers the user’s question from its first screen. Here are three examples, one per business model.
A SaaS vendor gets “What is a simple invoicing tool for a small team, without spreadsheets?”. That is a comparison intent, and the destination is not the product page but a comparison page: a table against the alternatives, the price shown, a free trial as the single CTA.
For a retailer, the prompt becomes “Which backpack fits as cabin baggage for a weekend with no checked bag?”, a purchase with a constraint. Instead of the homepage, the right target is an already filtered “cabin backpacks” category page, with airline dimensions, stock, price, delivery and reviews.
Then there is B2C lead gen, with “How much does attic insulation cost and what grants exist?”, a price intent. The installer does not send people to “Our services” but to a quote page: an explained range, the grants mentioned, a single form, the area covered.
In all three cases the page does not open with the brand. It opens with the question. The ad group’s context hints describe these same conversations, as sentences: “founder comparing simple invoicing software for a small team”. The context hints generator puts them in the expected format.
How do you align ad copy and H1?
By writing the three together: the title (up to 24 characters), the body (up to 48) and the page H1. The title names, the body adds information, the H1 rephrases the question.
| Title (≤ 24) | Body (≤ 48) | Page H1 |
|---|---|---|
| Invoicing for SMBs | Quotes, invoices, reminders, no spreadsheet | Simple invoicing for small teams |
| Cabin backpacks in stock | Airline sizes, fast delivery, free returns | The backpacks that fit in the cabin |
| Attic insulation quote | Online estimate, grants included | How much does attic insulation cost? |
The test is done out loud: read the title, then the body, then the H1. If the third line surprises you, message match is broken; if it feels like it repeats the other two, you are fine, since the user should have nothing to relearn on arrival.
Two mistakes come up often: a title that promises a price above a page that hides it behind a form, and a body that says “compare” above a page with no table. The native report shows both as clicks without conversions, and both are fixed on the page, not in Ads Manager.
If the campaign does not exist yet: create your first ChatGPT Ads campaign. As for the pages themselves, we do not build them; we say what they must contain, and we test them.