---
title: "Ads in AI Answers by 2027: Six Predictions You Can Hold Me To"
description: "Six dated predictions for how ads in ChatGPT and Google AI answers will be matched, measured and priced by 2027—and what to fix in your account now."
url: "https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/da496599-9a85-4953-935a-5cd45eb4d1ab.png"
published: "2026-11-03T06:51:50.556Z"
modified: "2026-11-03T06:51:50.616Z"
---

November 3, 2026 · 10 min read

# Ads in AI Answers by 2027: Six Predictions You Can Hold Me To

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

![A printed keyword list served on a china dinner plate with knife, fork and a place card dated Dec 31, 2027, ready to be eaten if the predictions fail.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/da496599-9a85-4953-935a-5cd45eb4d1ab.png)

In this article

1. [How the keyword lost control: four turning points](#how-the-keyword-lost-control-four-turning-points)
2. [1. By December 31, 2027, context will decide most AI-answer matches](#1-by-december-31-2027-context-will-decide-most-ai-answer-matches)
3. [2. By June 30, 2027, conversion signals will split winners from donors](#2-by-june-30-2027-conversion-signals-will-split-winners-from-donors)
4. [3. By September 30, 2027, the landing page will act like a bidding input](#3-by-september-30-2027-the-landing-page-will-act-like-a-bidding-input)
5. [4. By mid-2027, percent-of-spend pricing will cost agencies placements](#4-by-mid-2027-percent-of-spend-pricing-will-cost-agencies-placements)
6. [5. Through December 31, 2027, AI-ad reporting will stay thinner than Search](#5-through-december-31-2027-ai-ad-reporting-will-stay-thinner-than-search)
7. [6. Through December 31, 2027, early self-serve buyers will beat late buyers on cost](#6-through-december-31-2027-early-self-serve-buyers-will-beat-late-buyers-on-cost)
8. [What to change in your account this quarter](#what-to-change-in-your-account-this-quarter)

By December 31, 2027, your keyword list will not decide whether your ad shows inside ChatGPT or Google AI Mode. **Conversation context and the quality of your conversion data will.**

I spent most of the last decade building match-type structures by hand and mining search terms at 1am. I used to tell clients tight exact match was control. I was wrong about where control was moving. The machine has been taking query choice away from us for years; ads inside AI answers extend that shift. Here are six dated calls on what replaces the keyword list, and what would prove each one wrong.

## How the keyword lost control: four turning points

### 2018–July 2024: Matching moved beyond the words you picked

Google started loosening exact match with close variants around 2018, retired broad-match modifier into phrase in 2021, then made broad the default for new Search campaigns with Smart Bidding in [July 2024](https://hustlemarketers.com/). Each change gave the system more room to choose a query than the account structure appeared to allow.

That broke a useful old assumption: a tidy match-type build did not guarantee tidy matching. It also made conversion data more important. Broad plus Smart Bidding averages a cited 27% more conversions at 19% lower CPA than exact plus manual, but only with 30 to 50 conversions in 30 days, Enhanced Conversions and maintained negatives in place ([Optmyzr](https://optmyzr.com/)). Clean signals let the machine choose better queries. Dirty signals let it choose expensive ones.

### 2025: AI answers became ad inventory, not a separate account

AI Overviews rolled out on US mobile first, then desktop, and reached Australia in late 2025. AI Mode began as a US test in May 2025. Google did not hand advertisers a separate AI campaign to manage: ads inside AI Overviews and AI Mode come from existing Search, Shopping and Performance Max campaigns, with no separate campaign, no AI-only targeting option and no opt-out, according to [current setup guides](https://tofuads.com/).

The practical change was eligibility without a clean new control panel. More places for an ad to appear did not mean more ways to isolate where it appeared. That matters when you try to read a test rather than admire a blended report.

### February–May 2026: ChatGPT opened an auction without a keyword box

A US pilot went live February 9, 2026, for logged-in adults on Free and low-cost Go plans; Plus, Pro and Business saw no ads. Self-serve Ads Manager opened May 5, with Expedia among the early buyers ([pilot recap](https://trymaas.com/)). Advertisers write [context hints at ad-group level](https://contexthints.com/): plain-language descriptions of conversations where a product fits. OpenAI says those hints guide matching but do not act as exact-match rules.

The auction uses conversation context, ad content, the landing page and those hints. Past-chat personalization applies only when a user opts in. If your mental model is still bids on words, read how [ChatGPT ads get matched](https://groas.com/post/there-s-no-keyword-to-bid-on-how-ads-inside-chatgpt-actually) without one. The control you can exercise is in the business context you supply.

### July–October 2026: Google widened matching and kept reporting blended

By July 2026, only about 71% of exact-match impressions were truly exact in a study of 383 million EMEA ecommerce impressions; the rest came via AI Max matching. I watched old exact builds leak like this in real accounts. The structure still looked tidy. The matching was not.

Campaign-level broad and auto-created assets were auto-upgraded to [AI Max during September 1–30, 2026, with no opt-out](https://hustlemarketers.com/). Earlier, an Optmyzr analysis of 30,000 accounts in February 2026 had found exact lost about 10 points of spend share since 2022 while broad became dominant by budget. Until September, only broad match with Smart Bidding, AI Max, Shopping or Performance Max could serve in AI Mode, apart from explicit-intent queries. On [September 4, Google began a small experiment](https://ppc.land/) letting exact and phrase serve on those queries, while its help documents still said broad or keywordless AI Max was required for AI Overviews.

ChatGPT coverage grew to 60 to 89 countries, including the EEA, from August 24. AI Mode expanded on October 28 to 10 countries, including Australia, Canada and India, with multiturn follow-up ads in the US. Yet Google still folds AI ad results into standard campaign reports as Top Ads: triggering queries appear in aggregate, low-volume queries are omitted, and there is no AI Overviews-versus-AI Mode segment ([details here](https://cryptobriefing.com/)). **More inventory, less visibility** is the present-day trade. Here is where I think that line goes next.

## 1. By December 31, 2027, context will decide most AI-answer matches

A chat can hold five or ten turns of intent, budget, location and hesitation. A keyword holds two or three words. An auction that can use the conversation has a richer description of what the buyer needs, so the advertiser’s job shifts from collecting words to describing where the offer fits.

The evidence is already in the controls. ChatGPT Ads Manager has no keyword box; its [context hints](https://contexthints.com/) guide matching without acting as exact-match rules. On Google, broad plus Smart Bidding and AI Max already do much of the query choice. I stopped treating [keyword research and expansion as growth work](https://groas.com/post/keyword-research-tools-slowing-down-google-ads-2026) two years ago. It is maintenance now. Context is the growth lever.

**What proves me wrong:** by the end of 2027, either platform offers true keyword-level bidding for AI answers, reports the terms back against those keywords, and lets advertisers add exact negatives that reliably block them. If that happens, I will print this section and eat the paper.

For now, rewrite one ad group as five buyer situations in plain sentences, not 200 keywords. If you cannot describe the conversation where you win, you cannot expect a matching system to infer it from a list.

## 2. By June 30, 2027, conversion signals will split winners from donors

Loose matching moves the hard work from picking queries to predicting value. When a system can consider your ad for a thousand related conversations, it needs a reliable read on which ones turn into money. That read comes from Enhanced Conversions, offline imports and clean primary actions. Accounts with 30 to 50 real conversions in 30 days and deduplicated tracking give Smart Bidding something to learn from. Double-counted form fills and missing phone sales give it noise.

The [Optmyzr](https://optmyzr.com/) comparison above makes the condition hard to ignore: **27% more conversions at 19% lower CPA** for broad plus Smart Bidding versus exact plus manual came with signal requirements. I expect AI-answer auctions to make that difference more consequential, not less. I used to tell clients structure fixed performance. I was wrong. Structure sets the table; signal decides dinner.

**What proves me wrong:** by mid-2027, accounts with broken tracking and fewer than 10 conversions a month consistently win AI-answer placements at equal or lower CPA than clean-feed competitors in the same category.

Before raising an AI-ad budget by a dollar, fix one primary conversion action and its Enhanced Conversions wiring. Stop teaching the system that every form fill is a sale.

## 3. By September 30, 2027, the landing page will act like a bidding input

In ChatGPT’s matching, the landing page sits alongside the conversation, ad and [context hints](https://contexthints.com/). It is not merely the brochure someone sees after a click. A generic homepage that talks to everyone supplies less evidence of fit than a page answering a buyer’s specific question about price, service area, timeline or proof.

I learned the older version of this lesson watching Quality Score tank because a client swapped a landing page without telling me. This version has a stricter teacher. What the deck calls _post-click experience_, I call whether the page finishes the sentence the chat started.

**What proves me wrong:** by next fall, generic homepages hold equal impression share and CPA against situation-specific pages in controlled AI-placement tests.

Build a page for each buyer situation you target, with the offer and proof above the fold. Then test it against the generic page. Do not assume the click is the first time the system considers what that page says.

![A shopper talking to a chatbot while a specific and a generic landing page compete](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/417e26c2-a5ab-410e-b9ff-a03f434872b9.png)

## 4. By mid-2027, percent-of-spend pricing will cost agencies placements

Percent-of-spend deals still run 10% to 20% with $1,000 to $1,500 minimums, while some surveys put the range at 12% to 30% ([pricing guides](https://coseom.com/)). The fee rises with spend whether the extra spend produces pipeline or not. That model paid for human hours pulling levers in 2015. AI-answer placements demand a different daily job: rewriting context hints, supplying clean product and service detail, building specific pages and keeping conversion tracking intact.

A spend-based fee does not ensure any of that work gets done. It does reward a higher budget. Say you spend $20,000 a month. At 15%, you pay $3,000 whether CPA falls 20% or climbs 20%. I have no objection to paying for management; I object to a bill that treats more auction spend as evidence of more value.

**What proves me wrong:** by mid-2027, percent-of-spend shops publish AI-placement results that beat flat-fee operators on pipeline per dollar at equal spend.

Ask what work gets done each day and what it earns, then pay for that. A percentage on the media bill is not a work plan.

![An auction paddle beside a growing stack of coins while results stay flat](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/1c3bf1a4-6dd8-4ea6-a5ac-8871729cacb1.png)

## 5. Through December 31, 2027, AI-ad reporting will stay thinner than Search

A conversation is not a discrete query string you can print in a row, and OpenAI does not expose chats to advertisers. Its Ads Manager has [no search-term, placement or conversation report](https://contexthints.com/). Advertisers see aggregated impressions, clicks, spend, CTR, average CPC, CPM and conversions. Untagged clicks fall into Direct in analytics; phone leads need click-ID passthrough to tie back at all.

Google points in the same direction. AI ad results are folded into standard campaign reports as Top Ads, with triggering queries shown in aggregate, low-volume queries omitted and no segment separating AI Overviews from AI Mode ([breakdown here](https://cryptobriefing.com/)). Less detail about the match means more of the test has to rest on outcomes you can track yourself.

I miss search terms more than most. That report paid my rent for years. Waiting for it to come back is still the wrong bet.

**What proves me wrong:** by the end of 2027, either platform ships a prompt- or query-level export for AI-answer placements that ties to conversions and supports true negatives.

Tag every AI-ad click with its own UTM and use a dedicated conversion action. **If you cannot separate the click, you cannot read the test.**

## 6. Through December 31, 2027, early self-serve buyers will beat late buyers on cost

The pilot is the warning against calling _any_ early buyer cheap. ChatGPT Ads reportedly opened at a $60 fixed CPM with a $200,000 minimum in February 2026. The minimum fell to $50,000 in April, then disappeared after self-serve opened. Observed US CPMs now sit around $25 to $45, with CPC guidance of $3 to $5; software and finance CPCs ran $8 to $18 in the pilot. The India daily minimum is Rs725.

Those figures describe an auction changing shape, not a promise of cheap clicks forever. Scarcity pricing kept the pilot narrow. Self-serve made more conversations buyable and lowered the barrier to testing. My call is that advertisers who use this period to build conversion history will have a cost advantage over those who arrive after competition grows. The pilot buyers paid more to get in; the opportunity I mean is **early self-serve learning**, while the auction is still taking shape.

**What proves me wrong:** by the end of 2027, average AI-ad CPMs and CPCs in competitive categories sit flat or below 2026 levels despite three times the advertiser count.

Ring-fence a small, fixed test budget now, even if volume is low. Buy a readable test before you buy an expensive catch-up exercise.

## What to change in your account this quarter

These calls have loss conditions, not just dates. Keyword-level bids and reliable exact negatives would beat my first call. Broken tracking winning at equal CPA would beat my second. Generic pages holding their own would beat my third. I want those conditions on record because a prediction without one is just a mood.

If I am right, the work moves from lists to inputs. Pick one service line, say $5,000 to $20,000 in monthly spend, and make it the test bed:

1. **Fix the signal first.** Choose one primary conversion, wire up Enhanced Conversions and stop counting soft fills as sales.
2. Consolidate the Google test to broad plus Smart Bidding or AI Max so it stays eligible for AI Overviews and AI Mode.
3. Describe five to eight buying situations in hints and assets rather than expanding a 200-keyword list.
4. Ship a matching page for each situation. Measure CPA by situation, not by keyword, and tag AI-ad clicks so Direct does not swallow the read.
5. Set a small, fixed ChatGPT test budget. Pay for the daily work on context, tracking and pages, not for a percentage of what the auction takes.

This is why I run client work on [groas](https://groas.com/pricing): Search Ads covers Google, Bing and ChatGPT for one flat fee, with bids and pages worked around the clock. Context, signal and pages compound. Keyword lists do not. Start there.

## Frequently Asked Questions

### Will keywords still matter for ads in ChatGPT and Google AI answers by 2027?

The article predicts that by December 31, 2027, conversation context and conversion-data quality, not keyword lists, will decide most AI-answer matches. ChatGPT Ads Manager has no keyword box, and on Google, broad match plus Smart Bidding and AI Max already make most query choices. The recommended shift is describing five buyer situations in plain sentences instead of maintaining 200 keywords.

### How many conversions do I need per month for Smart Bidding to work well?

The article cites a threshold of 30 to 50 real conversions in 30 days with deduplicated tracking before broad match plus Smart Bidding can perform well. That Optmyzr comparison found 27% more conversions at 19% lower CPA versus exact match with manual bidding. Double-counted form fills and missing phone sales add noise that hurts the system's learning.

### Does my landing page affect whether my ad shows in AI answers?

Yes, according to the article's prediction that by September 30, 2027 the landing page will act like a bidding input. In ChatGPT's matching, the landing page sits alongside the conversation, ad content and context hints as evidence of fit. The advice is to build a page for each buyer situation, with offer and proof above the fold, and test it against a generic homepage.

### Is percent-of-spend pricing a bad way to pay a PPC agency?

The article argues percent-of-spend fees, typically 10% to 20% with $1,000 to $1,500 minimums, reward a higher budget rather than the daily work AI-answer placements require. That work includes rewriting context hints, supplying clean product detail, building specific pages and keeping tracking intact. The recommendation is to pay for defined daily work and results instead of a percentage of the media bill.

### Can I see search terms or conversation reports for ChatGPT ads?

No. OpenAI does not expose chats to advertisers, and its Ads Manager has no search-term, placement or conversation report; only aggregated impressions, clicks, spend, CTR, average CPC, CPM and conversions are shown. Google likewise folds AI ad results into standard campaign reports as Top Ads, with triggering queries shown only in aggregate. The article advises tagging every AI-ad click with its own UTM and a dedicated conversion action.

### Is now a good time to start advertising in ChatGPT answers?

The article predicts advertisers who build conversion history during the early self-serve period will have a cost advantage over those who arrive after competition grows. Observed US CPMs currently sit around $25 to $45 with CPC guidance of $3 to $5, after the pilot's $200,000 minimum was removed. The suggestion is to ring-fence a small, fixed test budget now rather than pay for an expensive catch-up later.

### Do I need a separate campaign to advertise in Google AI Overviews and AI Mode?

No. Ads inside AI Overviews and AI Mode come from existing Search, Shopping and Performance Max campaigns, with no separate campaign, no AI-only targeting option and no opt-out. Through most of 2026, only broad match with Smart Bidding, AI Max, Shopping or Performance Max could serve in AI Mode, apart from explicit-intent queries.

### How does targeting work in ChatGPT's Ads Manager?

There is no keyword box. Advertisers write context hints at ad-group level, which are plain-language descriptions of the conversations where a product fits. OpenAI says these hints guide matching but do not act as exact-match rules, and the auction also uses the conversation context, ad content and landing page.

## 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/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to#webpage","url":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to","name":"Ads in AI Answers by 2027: Six Predictions You Can Hold Me To","description":"Six dated predictions for how ads in ChatGPT and Google AI answers will be matched, measured and priced by 2027—and what to fix in your account now.","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/da496599-9a85-4953-935a-5cd45eb4d1ab.png"},"breadcrumb":{"@id":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to#breadcrumb"},"dateModified":"2026-11-03T06:51:50.616Z","mainEntity":{"@id":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to#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":"Ads in AI Answers by 2027: Six Predictions You Can Hold Me To","item":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to#article","url":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to","mainEntityOfPage":{"@id":"https://groas.com/post/ads-in-llms-by-2027-six-predictions-you-can-hold-me-to#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Ads in AI Answers by 2027: Six Predictions You Can Hold Me To","description":"Six dated predictions for how ads in ChatGPT and Google AI answers will be matched, measured and priced by 2027—and what to fix in your account now.","datePublished":"2026-11-03T06:51:50.556Z","dateModified":"2026-11-03T06:51:50.616Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/da496599-9a85-4953-935a-5cd45eb4d1ab.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"ads, 2027, answers, six, google, predictions, matching, conversions","wordCount":2193,"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-by-2027-six-predictions-you-can-hold-me-to#faq","mainEntity":[{"@type":"Question","name":"Will keywords still matter for ads in ChatGPT and Google AI answers by 2027?","acceptedAnswer":{"@type":"Answer","text":"The article predicts that by December 31, 2027, conversation context and conversion-data quality, not keyword lists, will decide most AI-answer matches. ChatGPT Ads Manager has no keyword box, and on Google, broad match plus Smart Bidding and AI Max already make most query choices. The recommended shift is describing five buyer situations in plain sentences instead of maintaining 200 keywords."}},{"@type":"Question","name":"How many conversions do I need per month for Smart Bidding to work well?","acceptedAnswer":{"@type":"Answer","text":"The article cites a threshold of 30 to 50 real conversions in 30 days with deduplicated tracking before broad match plus Smart Bidding can perform well. That Optmyzr comparison found 27% more conversions at 19% lower CPA versus exact match with manual bidding. Double-counted form fills and missing phone sales add noise that hurts the system's learning."}},{"@type":"Question","name":"Does my landing page affect whether my ad shows in AI answers?","acceptedAnswer":{"@type":"Answer","text":"Yes, according to the article's prediction that by September 30, 2027 the landing page will act like a bidding input. In ChatGPT's matching, the landing page sits alongside the conversation, ad content and context hints as evidence of fit. The advice is to build a page for each buyer situation, with offer and proof above the fold, and test it against a generic homepage."}},{"@type":"Question","name":"Is percent-of-spend pricing a bad way to pay a PPC agency?","acceptedAnswer":{"@type":"Answer","text":"The article argues percent-of-spend fees, typically 10% to 20% with $1,000 to $1,500 minimums, reward a higher budget rather than the daily work AI-answer placements require. That work includes rewriting context hints, supplying clean product detail, building specific pages and keeping tracking intact. The recommendation is to pay for defined daily work and results instead of a percentage of the media bill."}},{"@type":"Question","name":"Can I see search terms or conversation reports for ChatGPT ads?","acceptedAnswer":{"@type":"Answer","text":"No. OpenAI does not expose chats to advertisers, and its Ads Manager has no search-term, placement or conversation report; only aggregated impressions, clicks, spend, CTR, average CPC, CPM and conversions are shown. Google likewise folds AI ad results into standard campaign reports as Top Ads, with triggering queries shown only in aggregate. The article advises tagging every AI-ad click with its own UTM and a dedicated conversion action."}},{"@type":"Question","name":"Is now a good time to start advertising in ChatGPT answers?","acceptedAnswer":{"@type":"Answer","text":"The article predicts advertisers who build conversion history during the early self-serve period will have a cost advantage over those who arrive after competition grows. Observed US CPMs currently sit around $25 to $45 with CPC guidance of $3 to $5, after the pilot's $200,000 minimum was removed. The suggestion is to ring-fence a small, fixed test budget now rather than pay for an expensive catch-up later."}},{"@type":"Question","name":"Do I need a separate campaign to advertise in Google AI Overviews and AI Mode?","acceptedAnswer":{"@type":"Answer","text":"No. Ads inside AI Overviews and AI Mode come from existing Search, Shopping and Performance Max campaigns, with no separate campaign, no AI-only targeting option and no opt-out. Through most of 2026, only broad match with Smart Bidding, AI Max, Shopping or Performance Max could serve in AI Mode, apart from explicit-intent queries."}},{"@type":"Question","name":"How does targeting work in ChatGPT's Ads Manager?","acceptedAnswer":{"@type":"Answer","text":"There is no keyword box. Advertisers write context hints at ad-group level, which are plain-language descriptions of the conversations where a product fits. OpenAI says these hints guide matching but do not act as exact-match rules, and the auction also uses the conversation context, ad content and landing page."}}]}]}
```