---
title: "ChatGPT Ads Can’t Buy the Answer. They Have to Earn the Next Step."
description: "ChatGPT keeps ads separate from its answers. Here’s how an ad gets selected, what advertisers can measure, and why the offer and landing page matter more than clever copy."
url: "https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/200b7476-72bb-4ffc-ab92-f59b4a1267e2.png"
published: "2026-11-04T06:40:01.483Z"
modified: "2026-11-04T06:40:01.543Z"
---

November 4, 2026 · 9 min read

# ChatGPT Ads Can’t Buy the Answer. They Have to Earn the Next Step.

[](https://groas.com/author/alexander-perelman)

[Alexander Perelman](https://groas.com/author/alexander-perelman)Personally Tested

Head Of Product @ groas

[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![A porcelain speech bubble sealed under a glass dome as a roll of cash presses against it, with an open wooden step in front: ads can't buy the answer.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/200b7476-72bb-4ffc-ab92-f59b4a1267e2.png)

In this article

1. [First principle: the answer is not for sale](#first-principle-the-answer-is-not-for-sale)
2. [Layer one: separate systems protect the answer](#layer-one-separate-systems-protect-the-answer)
3. [Layer two: the conversation replaces the keyword list](#layer-two-the-conversation-replaces-the-keyword-list)
4. [Layer three: relevance changes the auction math](#layer-three-relevance-changes-the-auction-math)
5. [Layer four: measure the job, not just the click](#layer-four-measure-the-job-not-just-the-click)
6. [The landing page has to finish the conversation](#the-landing-page-has-to-finish-the-conversation)
7. [Apply the model to a buyer we have not met](#apply-the-model-to-a-buyer-we-have-not-met)

A home services client spending around $20k a month on Google asked me if he could pay to get ChatGPT to recommend his business in an answer. No. **The answer is not the ad slot**, and that distinction changes what he should spend money on.

OpenAI says the system that picks ads runs separately from the system that writes answers, with its own models and code. As ChatGPT writes a response, [the ads system checks for a relevant ad to show below it](https://www.searchenginejournal.com/openai-details-chatgpt-llm-ad-stack/593857/). Selecting an ad does not change the response. If the ad system is slow, the answer still arrives on time.

Everything else follows from that separation. **What you buy is the next step after the model has formed the buyer’s opinion.** The model tells someone with a leaking water heater what to do. Your ad may sit underneath that advice, fenced off from it. Clever copy cannot talk its way into the answer. Your offer and landing page have to make the next step worth taking.

## First principle: the answer is not for sale

Years of Google Search can give an advertiser a simple mental model: money helps move position. A higher bid and a better Quality Score can put an ad higher on the page, where it has a better shot at the click. I managed accounts with that model for the better part of a decade. It is easy to carry it into ChatGPT and assume that paying enough will make the model name you.

That trade does not exist here. The answer is produced separately from the ad. The ad appears as [a labeled sponsored unit below the response](https://www.digitbin.com/chatgpt-ads-how-ads-are-picked/), visually apart from it. A model-written intro line stays separate from the advertiser’s headline, description and CTA. On many turns, no ad appears at all. That is normal operation, not a placement problem to solve with a higher bid.

Keep one conversation in mind. At 11pm, a homeowner types: _My water heater is leaking. What should I do?_ ChatGPT gives its advice first: shut off the water and power to the unit, then get help quickly. Your ad for same-day water heater replacement may appear underneath. **You did not shape the advice. You can only offer a useful response to the need it exposed.**

## Layer one: separate systems protect the answer

Think of two rooms with a one-way window. ChatGPT writes the answer in one; the ad system uses the conversation’s context to decide whether an ad fits below it. Information does not flow back from an advertiser’s bid into the answer. The analogy breaks if you picture the ad system editing a completed draft through the glass. It does not. The useful part is the boundary: a late or empty ad call cannot hold up the response.

That boundary matters because trust is the inventory. Perplexity tried a different approach with [sponsored follow-up questions in late 2024](https://rankmehigher.co/learn/ads-in-llms/), priced around GBP30 to 60 CPM, then stopped taking advertisers for the format by early 2026. In practitioner threads, [68% of commenters reacted negatively to ChatGPT ads](https://www.wordstream.com/blog/chatgpt-ads), with trust loss the leading worry. Those comments are not a forecast for every reader. They do show the problem a platform creates if people suspect its advice can be bought.

For the water heater advertiser, the boundary is a constraint worth accepting. If the homeowner trusts the shutoff advice, the next question is whom to call. **Build for that question instead of trying to purchase the first answer.**

## Layer two: the conversation replaces the keyword list

In this system, you do not bid on _emergency plumber_ as a search keyword. You write [plain-language context hints describing conversations where your product helps](https://www.wordstream.com/blog/chatgpt-ads). OpenAI matches those hints using semantic embeddings alongside landing-page relevance, past performance and predicted engagement. The system is looking at the meaning of the conversation, not just a string of words.

I used to tell clients that match types were the control layer. I was wrong about how long that would remain true. For our example, a useful hint would describe a homeowner whose water heater is leaking and who needs replacement help outside business hours. A broad hint about home improvement reaches further but says much less about whether the advertiser can solve this problem now.

The matching process has two stages. A custom post-trained model called the _orchestrator_ condenses the conversation into summaries, rewrites them as retrieval queries and turns those queries into embeddings. It retrieves, filters and scores ad candidates. Then [a separate auction tool handles final ranking, pricing and selection](https://www.searchenginejournal.com/openai-details-chatgpt-llm-ad-stack/593857/). The orchestrator nominates; it does not choose the winner or set the price.

That is why the hint and the destination page need to tell the same story. A hint about urgent water heater replacement paired with a generic services page gives the system, and eventually the homeowner, less reason to believe the fit is real. **Write the hint like a dispatcher describing the call, then make sure the page answers it.**

## Layer three: relevance changes the auction math

Candidates that survive the orchestrator enter [a relevance-weighted second-price auction](https://www.thekeyword.co/learn/chatgpt-ads). The winner pays just above the next-highest bid, and a lower bid with stronger relevance can beat a higher one. The stated payment models are CPM, with a $60 default maximum per 1,000 impressions, and CPC, with a suggested starting maximum of $3 to $5. CPA payment on defined actions is in pilot and requires Pixel and Conversions API tracking.

For the water heater advertiser, a $5 click bid paired with a vague hint and a slow, generic page can lose to a $3.20 bid tied to a relevant conversation and a page about same-day replacement. The lower bidder does not win by magic. Relevance is part of the ranking, and the page helps establish whether the ad belongs in that conversation.

Even an auction win does not guarantee an appearance. OpenAI applies [a final contextual suitability check and a delivery-window check](https://www.searchenginejournal.com/openai-details-chatgpt-llm-ad-stack/593857/). An ad that arrives after the answer completes is dropped. Low delivery could reflect relevance, targeting, policy, budget or timing. OpenAI logs those stages internally but has not said how much of that detail advertisers can see.

Nor does an appearance guarantee attention. Early testers have reported around 0.5% CTR and $10 CPMs in one account, and five-second visits from ChatGPT clicks in another. Reach is also limited to the Free and Go tiers; Plus, Pro, Business and Enterprise are ad-free. Those observations are reasons to test the entire path to a customer, not to treat a cheap-looking impression as a result. **Bid for conversations your offer can close.**

## Layer four: measure the job, not just the click

Measurement depends on three connected pieces:

1. Install the [oaiq JavaScript Pixel](https://www.kioksy.se/articles/chatgpt-ads-measurement-what-openai-announced-and-how-to-read-it.html) on your site.
2. Send the same events server-side through the Conversions API, using a shared event ID and the _oppref_ click reference for deduplication.
3. Keep UTMs so your own analytics can follow the visit.

Reporting offers 7-, 14- or 30-day click windows and an optional zero- or one-day view window, using last-touch attribution. OpenAI’s October measurement update named more than 10 attribution and lift partners, but its FAQ still says there are no stable benchmarks across advertisers or industries. Set up the Pixel and CAPI pair before spending; otherwise, you lose a crucial view of what happened after the click.

You still will not see everything I would want from a search account. In Google, I could open a search term report and learn from the queries. Here, OpenAI records retrieval, filtering, auction and suitability stages internally without saying how much advertisers get to inspect. A missed impression could mean poor relevance, a filter, an exhausted budget or an ad that arrived too late. The reporting window creates another blind spot: someone sees the water heater ad, closes the chat, searches the brand the next morning and books. Last-touch reporting may not give the first encounter credit.

That does not make measurement pointless. It tells you where to put your attention. **Judge the test on booked jobs and pipeline, not platform CTR alone.**

## The landing page has to finish the conversation

The answer has already done part of the buyer’s thinking. For our homeowner, it has established urgency and the need for help. An ad that merely announces _Water Heater Services_ adds little. A page that addresses a leaking unit, explains the next step, gives a price range and puts the phone number and booking button where a stressed homeowner can find them has a job to do.

This is where I would spend more effort than I would on a clever headline. The ad has limited room, and it cannot revise what the model said. The landing page can show whether the advertiser’s offer fits the situation the model described. If that page is slow or generic, even a well-matched click lands nowhere useful. **Copy cannot rescue a page that ignores what the buyer has just learned.**

The same logic changes the management work. Hints, pages, bids and Pixel plus CAPI signals need to be adjusted together from live delivery. A consultant who audits the hints once and sends a PDF is not doing the same job as someone who changes the campaign and its destination when the evidence calls for it. The weekly work is concrete:

- Rewrite context hints around the conversations the offer can serve.
- Test the page the ad points to, not just its headline.
- Tune bids and budgets alongside relevance and delivery.
- Hold one person accountable for booked jobs, not impressions.

That is the case for [autonomous management rather than periodic account reviews](https://groas.com/post/autonomous-google-ads-management-2026-agencies-in-house-teams-budget). groas offers [Search Ads across Google, Bing and ChatGPT for a flat $1,499 a month](https://groas.com/pricing), with bids, keywords and budgets worked around the clock and a landing page for every ad group. The point is not that every advertiser needs the same operating model. It is that this channel asks the manager to act on the page as well as the ad. **If your manager cannot change the page, they cannot manage the whole path to a customer.**

## Apply the model to a buyer we have not met

Now take a B2B buyer three prompts into vendor research. She has pasted in her requirements, asked for tradeoffs between two platforms, and received an answer that makes implementation time more important than the feature she first cared about. Her priorities have shifted. An ad underneath cannot argue its way into the answer, and generic copy about an _all-in-one solution_ does not address the new concern.

Work through the layers. The answer stands apart from the ad. Her conversation supplies the context that may make an ad eligible. Relevance affects whether that ad reaches and wins the auction. Then the click lands somewhere you control. A page with implementation timelines, migration notes and a real price range meets the buyer where the answer left her. **The useful next step accepts her new priority instead of pretending she never changed her mind.**

![A printed AI answer above a separate sponsored card for water heater repair](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/6c378ad1-b41b-4c39-967a-13059c60d15a.png)

Take another case: a shopper comparing two similarly priced cordless vacuums. The answer weighs suction, battery life and pet hair handling, then favors one model on the feature he cares about most. If yours is the losing model, your ad cannot flip the comparison above it. A headline full of superlatives only makes the gap more obvious.

The model is the same, but the next step changes. A page that names both models, shows where yours fits a specific home and offers a 30-day trial gives the shopper a reason to reconsider without claiming the answer said something else. The ad earns consideration only if the offer makes that reconsideration worthwhile. My water heater client stopped asking how to buy the answer and started asking what his page would promise after it. That is the right question in both cases. **The answer belongs to the model. The next step belongs to whoever earns it.**

## Frequently Asked Questions

### Does paying more money guarantee a featured spot in a ChatGPT response?

ChatGPT writes answers separately from the system that selects ads below them. Therefore, high bids cannot manipulate the content of the model-generated advice. The ad appears as a labeled sponsored unit visually apart from the model's response.

### How should an advertiser describe their service to be found in a ChatGPT ad?

Advertisers must write plain-language context hints that describe specific conversations where their product is relevant. The system uses semantic embeddings to match the meaning of the conversation to the ad hint. The system is looking at the meaning of the conversation, not just a string of words.

### Can a lower bid still win a ChatGPT ad auction?

Candidates enter a relevance-weighted second-price auction, meaning a lower bid paired with high relevance can beat a higher bid. Strong relevance helps establish that the ad fits the specific context of the conversation. Relevance is part of the ranking process.

### What is required to measure the performance of ChatGPT ads?

Effective measurement requires installing the oaiq JavaScript Pixel and sending events server-side through the Conversions API. Using UTM parameters is also necessary for website analytics to track the visit. Setting up these tools before spending is crucial.

### What must a landing page do after an ad is clicked in ChatGPT?

The landing page must be designed to address the urgent need or specific scenario presented in the AI's advice. Because the ad cannot revise the model's content, the page must offer a clear, useful next step that completes the conversation.

## Run Your Marketing On groas

Plug ChatGPT, Claude or any AI agent into groas and let it do the work of a whole marketing team: $19 a month flat with every feature included, and autopilot whenever you’re ready.

[Start For $19](https://groas.com/pricing)

[Or have groas run it for you](https://groas.com/done-for-you)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","description":"groas is the marketing infrastructure for your own AI: it connects your AI to all your marketing and gives it proprietary marketing data collected 24/7. Plug Claude, ChatGPT or any AI agent into groas and run all your marketing from it, for a tiny fraction of the usual cost, with the groas app there whenever you want to see more. It is the layer ChatGPT, Claude and any AI assistant run a business’s online marketing on. Through groas an assistant can do anything you could do natively in Google Ads, Bing Ads, Meta Ads and ChatGPT ads, on your website and in AI search, and, where the owner switches it on, can run connected campaigns by itself, so AI agents do the work a whole marketing team used to. groas also gives those agents large-scale proprietary marketing data it collects around the clock, on any schedule the owner chooses: every ad competitors run on Google and Meta, their landing pages and full funnels with the offers and prices in them, Reddit customer research, AI search answers from ChatGPT, Google AI Overviews, Gemini, Claude and Perplexity, and SEO data from Semrush and Ahrefs, backlinks included ($19 a month, flat, with everything included, for unlimited teammates on one shared workspace and unlimited ad accounts and websites, delivered as an MCP server). Separately, its done-for-you service has senior people and proprietary quantitative systems run paid ads, SEO and AI search for businesses, and white label for agencies, from $5,000 a month.","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer#webpage","url":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer","name":"ChatGPT Ads Can’t Buy the Answer. They Have to Earn the Next Step.","description":"ChatGPT keeps ads separate from its answers. Here’s how an ad gets selected, what advertisers can measure, and why the offer and landing page matter more than clever copy.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/200b7476-72bb-4ffc-ab92-f59b4a1267e2.png"},"breadcrumb":{"@id":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer#breadcrumb"},"dateModified":"2026-11-04T06:40:01.543Z","mainEntity":{"@id":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"ChatGPT Ads Can’t Buy the Answer. They Have to Earn the Next Step.","item":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer#article","url":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer","mainEntityOfPage":{"@id":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"ChatGPT Ads Can’t Buy the Answer. They Have to Earn the Next Step.","description":"ChatGPT keeps ads separate from its answers. Here’s how an ad gets selected, what advertisers can measure, and why the offer and landing page matter more than clever copy.","datePublished":"2026-11-04T06:40:01.483Z","dateModified":"2026-11-04T06:40:01.543Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/200b7476-72bb-4ffc-ab92-f59b4a1267e2.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"answer, page, chatgpt, next step, ads, next, water heater, step","wordCount":1988,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"My background includes Goldman Sachs and Stanford Computer Science.","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/ads-in-llms-explained-why-the-ad-never-touches-the-answer#faq","mainEntity":[{"@type":"Question","name":"Does paying more money guarantee a featured spot in a ChatGPT response?","acceptedAnswer":{"@type":"Answer","text":"ChatGPT writes answers separately from the system that selects ads below them. Therefore, high bids cannot manipulate the content of the model-generated advice. The ad appears as a labeled sponsored unit visually apart from the model's response."}},{"@type":"Question","name":"How should an advertiser describe their service to be found in a ChatGPT ad?","acceptedAnswer":{"@type":"Answer","text":"Advertisers must write plain-language context hints that describe specific conversations where their product is relevant. The system uses semantic embeddings to match the meaning of the conversation to the ad hint. The system is looking at the meaning of the conversation, not just a string of words."}},{"@type":"Question","name":"Can a lower bid still win a ChatGPT ad auction?","acceptedAnswer":{"@type":"Answer","text":"Candidates enter a relevance-weighted second-price auction, meaning a lower bid paired with high relevance can beat a higher bid. Strong relevance helps establish that the ad fits the specific context of the conversation. Relevance is part of the ranking process."}},{"@type":"Question","name":"What is required to measure the performance of ChatGPT ads?","acceptedAnswer":{"@type":"Answer","text":"Effective measurement requires installing the oaiq JavaScript Pixel and sending events server-side through the Conversions API. Using UTM parameters is also necessary for website analytics to track the visit. Setting up these tools before spending is crucial."}},{"@type":"Question","name":"What must a landing page do after an ad is clicked in ChatGPT?","acceptedAnswer":{"@type":"Answer","text":"The landing page must be designed to address the urgent need or specific scenario presented in the AI's advice. Because the ad cannot revise the model's content, the page must offer a clear, useful next step that completes the conversation."}}]}]}
```
