---
title: "There Is No Search Term Report in a Chatbot"
description: "ChatGPT Ads match conversations, not keyword lists. Here is what replaces match types and query mining, and how to prepare pages, hints, and measurement."
url: "https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/a9241c7b-a934-4a0a-872e-d6791d31fb04.png"
published: "2026-11-03T06:50:05.812Z"
modified: "2026-11-03T06:50:05.884Z"
---

November 3, 2026 · 10 min read

# There Is No Search Term Report in a Chatbot

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![Tabletop model of a homeowner talking beside a green pool while an ad buyer behind the fence listens through a knothole, his clipboard blank.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/a9241c7b-a934-4a0a-872e-d6791d31fb04.png)

In this article

1. [Start with the four words Google can show you](#start-with-the-four-words-google-can-show-you)
2. [A chatbot has to interpret the conversation first](#a-chatbot-has-to-interpret-the-conversation-first)
3. [The session changes which ads fit](#the-session-changes-which-ads-fit)
4. [Give the model a page that answers the actual need](#give-the-model-a-page-that-answers-the-actual-need)
5. [Winning the auction is not the same as appearing](#winning-the-auction-is-not-the-same-as-appearing)
6. [Check who can see the placement at all](#check-who-can-see-the-placement-at-all)
7. [Know which parts of the system stay hidden](#know-which-parts-of-the-system-stay-hidden)
8. [Apply the model to the next assistant pitch](#apply-the-model-to-the-next-assistant-pitch)
9. [Prepare the offer before you spend](#prepare-the-offer-before-you-spend)

I spent nine years living inside the search term report. In ChatGPT Ads, the report I used to mine at 1am does not exist. Neither does the keyword list I would have tuned after reading it.

Take a homeowner with a green pool. In Google, he types `why is my pool green`. In a chatbot, he says the water turned after a storm, then that the filter runs fine, then that his kids are swimming this weekend. Those turns give an ad system more to work with than four typed words. They also change what an advertiser can control. **The match key is the conversation, not a keyword.**

## Start with the four words Google can show you

In Google, the familiar loop starts with a string. The homeowner types `why is my pool green`, and Google calls that the [search term](https://support.google.com/google-ads/answer/2472708?visit_id=638990853097926966-2507370668\&rd=1). What I add to the account, say `pool cleaning service`, is the keyword. Google matches the search term against eligible keywords, subject to match type, and ranks eligible ads by Ad Rank.

The [search term report is where I did most of my real work](https://support.google.com/google-ads/answer/2472708?visit_id=638990853097926966-2507370668\&rd=1). It showed which typed strings triggered my keyword, what match type fired, and what each cost. When junk slipped through, I added it as a negative. When a good variant appeared, I broke it out. Query in, bid out, report back.

That loop depends on being able to inspect the strings that brought people in. Take away the strings and the report, and you cannot recreate the loop by renaming your ad groups. **Before changing tactics, notice what the old tactic required.**

## A chatbot has to interpret the conversation first

In ChatGPT, there is no single string to bid on. OpenAI runs a [separate ad stack built around stateful sessions](https://onemetrik.com/market-insights/chatgpt-ad-stack/). A custom post-trained model, called an orchestrator, rewrites the conversation into commercial-intent queries. Retrieval, eligibility and relevance checks follow. Then an auction sets ranking and price; it does not write the answer.

That rewrite sits between the homeowner’s words and the ad. By the time he mentions the working filter and Saturday swim, the useful signal is no longer just `why is my pool green`. It is the situation he has described across turns. A filter-repair offer and a pool-vacuum offer may look similar beside the original four words. They look different beside the whole exchange.

[There is no keyword bidding on ChatGPT Ads](https://guptadeepak.com/guides/llm-ads-explained/). Instead, an advertiser supplies a context hint: a freeform natural-language description at ad-group level, capped at 280 characters. The system matches that hint against the live conversation, and the [auction weighs the hint alongside title, copy and landing page in a relevance-weighted second-price auction](https://guptadeepak.com/guides/llm-ads-explained/).

So I would describe the homeowner’s situation in the hint, not paste in the keyword I once bought. The phrase _pool vacuum_ need not appear in his chat for the offer to make sense. **Write for the need the conversation reveals, not the string you hope someone types.**

## The session changes which ads fit

Other AI surfaces show why the distinction matters, though they do not all use the same controls. Copilot matches on the [whole conversation within a single session, not just the last message](https://www.webfx.com/blog/ai/copilot-search-ads/), then shows ads beneath the organic response with a short explanation of why they fit. Google has two routes: [direct query matching when the search already has commercial intent, and contextual matching to both the query and the AI response when intent appears in the next steps](https://support.google.com/google-ads/answer/18826097?hl=en). Google’s example uses `why is my pool green` to surface pool-vacuum ads because the answer points that way. **Exact and phrase qualify only for the first route. Broad, AI Max, PMax and Shopping can qualify for both.**

I used to tell clients negatives were profit. That was sound advice for the account structure we were running, but I cannot carry it over unchanged. [ChatGPT has no match types, negative keyword lists or search term report](https://guptadeepak.com/guides/llm-ads-explained/). It has no dedicated exclusion control for a conversation type; a narrower hint is the available way to steer away from one. Port a 400-term negative list and you will find nowhere to paste it.

The practical difference is easy to miss. In Google, I could see an unwanted term and exclude it. Here, I have to describe more clearly which conversations belong to the offer. **A hint is a relevance instruction, not a negative-keyword list in prose.**

## Give the model a page that answers the actual need

A hint gets an offer into consideration. The destination has to support it. Relevance in the ChatGPT auction draws on [four inputs: context hint, landing page, ad title and ad copy](https://www.contexthints.com/services/chatgpt-ads-audit.html). Image is explicitly not one of them. The system reads the page against the conversation to decide whether the two belong together.

This is where one generic pool-service page becomes a problem. Our homeowner’s thread could lead to algae treatment, filter diagnosis or a same-week vacuum rental for the kids’ swim. Those are different outcomes. A page that says only “we handle all your pool needs” asks the system, and the homeowner, to fill in the useful details.

I would give each outcome its own page, with the details that let someone act: price or price range, service area and booking steps, as appropriate. For the vacuum rental, Saturday availability matters to this conversation. For filter diagnosis, the page needs to make that service clear even though the homeowner says the filter runs fine. **The page should make the offer legible without asking a model to guess.**

That does not make the page a literal bid, or guarantee a slot. The auction still ranks ads, and other checks still apply. But a clearer page gives the relevance checks something concrete to read. One answer also holds far fewer ads than a results page, so [inventory is scarce by design](https://guptadeepak.com/guides/llm-ads-explained/). I would rather compete with a page that plainly answers the situation than one that merely repeats a service name.

## Winning the auction is not the same as appearing

Even a relevant offer can disappear before the reader sees it. [Final contextual and timing checks can drop an auction winner; delivery can miss the response deadline, and ChatGPT will not slow its answer to wait for an ad](https://onemetrik.com/market-insights/chatgpt-ad-stack/). Answer generation and ad selection run on separate systems. Winning the auction math, in other words, does not guarantee an impression.

The reporting gives me less to debug than a search account would. [OpenAI Ads Manager has no search-term, placement, auction-insights or ad-level A/B report](https://www.contexthints.com/services/chatgpt-ads-audit.html). It provides impressions, clicks, spend, CPC and CPM, with conversions aggregated by design and cuts limited to Device and Country. I do not see the chat or the history that triggered an ad. Conversions take 24 to 48 hours to appear.

That changes the measurement job. I can inspect delivery and clearing price for clues about relevance, and pass [URL macros such as `{campaign_id}` and `{ad_id}`](https://www.contexthints.com/services/chatgpt-ads-audit.html) downstream so my analytics can distinguish traffic the platform reports more broadly. I cannot turn those IDs into a missing transcript or run a search-term cleanup afterward. **Tag what you can observe, then judge the traffic by the work it produces.**

The missing report is not a minor interface complaint. Without it, there is no query-mining loop to run. It is another reason [keyword-centric structure keeps failing](https://groas.com/post/keyword-centric-google-ads-strategy-failing-2026) as matching moves from strings toward interpreted intent.

## Check who can see the placement at all

Say you spend $20k a month chasing operations managers. Before debating hints, check the audience. [ChatGPT ads serve logged-in adults on the Free and Go tiers; Plus, Pro, Business, Enterprise and Education are ad-free](https://guptadeepak.com/guides/llm-ads-explained/). If the buyer you want uses a company-paid plan, this placement cannot reach that person there. A beautifully written page will not change eligibility.

Scarce answer-level inventory narrows the opportunity further. Copilot introduces a different constraint: it [auto-enrolls eligible search and PMax campaigns, with no opt-out and no guaranteed display](https://www.webfx.com/blog/ai/copilot-search-ads/). Those are not interchangeable buying experiences, even if both appear beside an AI response. **Eligibility is not delivery, and delivery is not a qualified customer.** That is the sequence I would check before moving budget.

## Know which parts of the system stay hidden

The broad shape of the OpenAI stack is visible; its tuning is not. The auction is [relevance-weighted second-price](https://guptadeepak.com/guides/llm-ads-explained/), and final checks can still drop a winner. But OpenAI has not published the weights for hint versus page versus copy, the orchestrator’s rewrite rules or the threshold that stops a winning ad at delivery. Reporting remains aggregated, conversions lag, and the advertiser gets no chat transcript.

I would not pretend a few clicks can reveal those missing rules. They may tell me whether an offer gets traction, especially when I can follow tagged visits to booked work. They cannot tell me exactly how a private conversation was rewritten.

There is a reason to keep that boundary in view. Perplexity [walked away from sponsored answers over answer-trust risk](https://guptadeepak.com/guides/llm-ads-explained/). In ChatGPT, the auction does not write the answer. Separating an ad from the answer protects the usefulness of the surface that carries it. **Build a test around the signals you receive, not the private chat you wish you could inspect.**

## Apply the model to the next assistant pitch

Every new assistant placement is going to sound, at first, like Google with a new skin. I would run the green-pool conversation through it instead. At turn two, the homeowner says the filter runs fine. A system matching only his first message has little reason to change its view. A system using the session has a reason to weigh filter repair differently from vacuum rental. That is the question to settle before carrying over an account structure.

**I would ask three questions before funding a test:**

1. **What is the match key: one message or the session?** Does the system read the history, and does it use its own answer, as [Google can use the query and AI response](https://support.google.com/google-ads/answer/18826097?hl=en)? [Copilot reads the whole session](https://www.webfx.com/blog/ai/copilot-search-ads/). Those distinctions tell me whether I am buying against a string or against developing context.
2. **What replaces the keyword, and can I exclude an unwanted fit?** ChatGPT supplies a 280-character hint and [no negative list](https://guptadeepak.com/guides/llm-ads-explained/). If a new surface also lacks exclusions, I need a narrower description of the situation and a page that supports it. I cannot plan to clean up irrelevant conversations in a report that does not exist.
3. **What can I measure after the ad runs?** If there is [no chat transcript and reporting offers only Device and Country cuts](https://www.contexthints.com/services/chatgpt-ads-audit.html), I cannot judge the test by how neatly I debug individual clicks. I need downstream cost per qualified lead or booked job, with enough time for conversions to appear.

That is a first pass, not a way to reverse-engineer a private auction. It tells me what to prepare, what to measure and which familiar PPC tools to leave behind.

## Prepare the offer before you spend

For the pool homeowner, I would prepare three distinct destinations: algae cleanup with a price range and service area; filter diagnosis with photos showing what a clog looks like; and emergency vacuum rental with Saturday availability and a booking button. Each page should let the reader understand the offer quickly, while giving the matching system plain text it can interpret. I would write each hint as a situation: who needs help, what has already happened and what a useful outcome looks like. **A strong hint describes the conversation you want; a strong page makes the offer clear when that conversation arrives.**

Then I would tag the ads with their IDs, allow for the reporting lag and judge results by cost per booked job or qualified pipeline. I used to walk into every new placement with a keyword list and a negative list. I would be wrong to bring them here as the plan. The match key changed, so the work changed: clearer offers, readable pages and tracking that survives aggregated reporting.

That is why we run [Google, Bing and ChatGPT as one Search Ads service at groas](https://groas.com/pricing) rather than selling chatbot ads as a separate toy. The channels do not share every control, but the customer’s intent can move across them. The offer and its destination need to make sense wherever that customer finds them.

One old habit still pays. I would run a small test rather than argue from an auction diagram: take a slice of that $20k monthly budget for two weeks, choose one situation, and try three hints that describe it slightly differently, each pointed at its own page. Watch which hint clears at a lower price and which page turns visits into booked work. There is no query-mining session afterward. Keep the page that closes; rewrite the hint that does not. **That is the optimization loop until the platforms show us more.**

## Frequently Asked Questions

### Why is there no search term report for ChatGPT Ads?

ChatGPT Ads are matched against the conversation as a whole, not against a typed search string or a keyword list. Because no single string triggers the ad, there is no search term report and no keyword list to tune. The match key is the chat session itself.

### What replaces keyword bidding in ChatGPT Ads?

You supply a context hint: a freeform natural-language description of the situation, capped at 280 characters, at ad-group level. The system matches that hint against the live conversation. There are no match types and no negative keyword lists; a narrower hint is the available way to steer away from unwanted conversations.

### Which inputs count toward relevance in the ChatGPT Ads auction?

Four inputs count: the context hint, the landing page, the ad title and the ad copy. Image is explicitly not one of them. The system reads the page against the conversation to decide whether the offer and the chat belong together.

### How should I set up landing pages for chatbot ads?

Give each distinct outcome its own page, with the details that let someone act, such as a price or price range, service area and booking steps. For example, a vacuum-rental page should show Saturday availability, and a filter-diagnosis page should make that service clear. Avoid one generic page that asks the system to guess the useful details.

### Does winning the ChatGPT ad auction guarantee my ad shows?

No. Final contextual and timing checks can drop an auction winner, and delivery can miss the response deadline because ChatGPT will not slow its answer to wait for an ad. Answer generation and ad selection run on separate systems, so winning the auction math does not guarantee an impression.

### What reporting is available in OpenAI Ads Manager?

OpenAI Ads Manager provides impressions, clicks, spend, CPC and CPM, with conversions aggregated by design and cuts limited to Device and Country. There is no search-term, placement, auction-insights or ad-level A/B report, and you cannot see the chat that triggered an ad. Conversions take 24 to 48 hours to appear.

### How can I track ChatGPT ad traffic without a transcript?

Tag the ads with URL macros such as {campaign_id} and {ad_id} and pass those IDs downstream so your analytics can distinguish the traffic the platform reports only broadly. You cannot recover a missing chat transcript or run a search-term cleanup afterward, so judge the traffic by downstream results such as cost per booked job or qualified lead.

### Who can actually see ads in ChatGPT?

ChatGPT ads serve logged-in adults on the Free and Go tiers. Plus, Pro, Business, Enterprise and Education plans are ad-free. If the buyer you want uses a company-paid plan, that placement cannot reach them there, no matter how well your page is written.

## Paid Ads & SEO, Run For You

Businesses pay one flat monthly fee per service, agencies resell it under their own brand, and groas does the work either way.

[See Pricing](https://groas.com/pricing)

## 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 a done-for-you marketing service, not software: a dedicated strategist and the groas team run search ads, Meta ads, SEO and AI search for businesses, and white label for agencies.","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/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms#webpage","url":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms","name":"There Is No Search Term Report in a Chatbot","description":"ChatGPT Ads match conversations, not keyword lists. Here is what replaces match types and query mining, and how to prepare pages, hints, and measurement.","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/a9241c7b-a934-4a0a-872e-d6791d31fb04.png"},"breadcrumb":{"@id":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms#breadcrumb"},"dateModified":"2026-11-03T06:50:05.884Z","mainEntity":{"@id":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms#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":"There Is No Search Term Report in a Chatbot","item":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms#article","url":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms","mainEntityOfPage":{"@id":"https://groas.com/post/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"There Is No Search Term Report in a Chatbot","description":"ChatGPT Ads match conversations, not keyword lists. Here is what replaces match types and query mining, and how to prepare pages, hints, and measurement.","datePublished":"2026-11-03T06:50:05.812Z","dateModified":"2026-11-03T06:50:05.884Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/a9241c7b-a934-4a0a-872e-d6791d31fb04.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"report, page, search term, search, term report, term, ads, chatgpt","wordCount":2155,"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/there-is-no-search-term-report-in-a-chatbot-how-ads-in-llms#faq","mainEntity":[{"@type":"Question","name":"Why is there no search term report for ChatGPT Ads?","acceptedAnswer":{"@type":"Answer","text":"ChatGPT Ads are matched against the conversation as a whole, not against a typed search string or a keyword list. Because no single string triggers the ad, there is no search term report and no keyword list to tune. The match key is the chat session itself."}},{"@type":"Question","name":"What replaces keyword bidding in ChatGPT Ads?","acceptedAnswer":{"@type":"Answer","text":"You supply a context hint: a freeform natural-language description of the situation, capped at 280 characters, at ad-group level. The system matches that hint against the live conversation. There are no match types and no negative keyword lists; a narrower hint is the available way to steer away from unwanted conversations."}},{"@type":"Question","name":"Which inputs count toward relevance in the ChatGPT Ads auction?","acceptedAnswer":{"@type":"Answer","text":"Four inputs count: the context hint, the landing page, the ad title and the ad copy. Image is explicitly not one of them. The system reads the page against the conversation to decide whether the offer and the chat belong together."}},{"@type":"Question","name":"How should I set up landing pages for chatbot ads?","acceptedAnswer":{"@type":"Answer","text":"Give each distinct outcome its own page, with the details that let someone act, such as a price or price range, service area and booking steps. For example, a vacuum-rental page should show Saturday availability, and a filter-diagnosis page should make that service clear. Avoid one generic page that asks the system to guess the useful details."}},{"@type":"Question","name":"Does winning the ChatGPT ad auction guarantee my ad shows?","acceptedAnswer":{"@type":"Answer","text":"No. Final contextual and timing checks can drop an auction winner, and delivery can miss the response deadline because ChatGPT will not slow its answer to wait for an ad. Answer generation and ad selection run on separate systems, so winning the auction math does not guarantee an impression."}},{"@type":"Question","name":"What reporting is available in OpenAI Ads Manager?","acceptedAnswer":{"@type":"Answer","text":"OpenAI Ads Manager provides impressions, clicks, spend, CPC and CPM, with conversions aggregated by design and cuts limited to Device and Country. There is no search-term, placement, auction-insights or ad-level A/B report, and you cannot see the chat that triggered an ad. Conversions take 24 to 48 hours to appear."}},{"@type":"Question","name":"How can I track ChatGPT ad traffic without a transcript?","acceptedAnswer":{"@type":"Answer","text":"Tag the ads with URL macros such as {campaign_id} and {ad_id} and pass those IDs downstream so your analytics can distinguish the traffic the platform reports only broadly. You cannot recover a missing chat transcript or run a search-term cleanup afterward, so judge the traffic by downstream results such as cost per booked job or qualified lead."}},{"@type":"Question","name":"Who can actually see ads in ChatGPT?","acceptedAnswer":{"@type":"Answer","text":"ChatGPT ads serve logged-in adults on the Free and Go tiers. Plus, Pro, Business, Enterprise and Education plans are ad-free. If the buyer you want uses a company-paid plan, that placement cannot reach them there, no matter how well your page is written."}}]}]}
```