Why keyword research fails on ChatGPT
Keyword research is the foundation of Google Ads. You identify short phrases people type into a search bar, bid on them, and show ads when those phrases appear. The problem is that ChatGPT conversations do not look anything like search queries.
A Google search looks like this: “best CRM small business.” Four words. A ChatGPT prompt covering the same need looks like this: “I run a 15-person marketing agency and we need a CRM that integrates with our existing HubSpot setup, handles client-facing reporting, and costs less than $80 per seat. We also need mobile access for our field reps. What should we consider?”
No single keyword captures the intent embedded in that paragraph. The user has described their company size, industry, integration requirements, budget ceiling, feature needs, and use case in a single message. On Google, an advertiser would need to build keyword lists covering dozens of variations. On ChatGPT, there are no keywords to bid on at all.
This is not a limitation of ChatGPT’s ad system. It is a reflection of how people communicate with conversational AI. When users talk to ChatGPT, they write in natural language. They describe problems in full sentences and paragraphs, not in two-to-five-word fragments. Intent is embedded in paragraphs, not search queries.
The privacy landscape reinforces this shift. Over 75% of global web traffic now occurs in cookie-limited environments (Google Privacy Sandbox Report, 2025). Safari and Firefox block third-party cookies by default. Google Chrome is deprecating them. Behavioral targeting that relies on tracking users across websites is becoming less effective every quarter. Contextual targeting, which analyzes the content a user is engaged with rather than who the user is, is the approach that scales in this environment.
ChatGPT’s ad system was built from the ground up for this reality. It does not track users across the web. It does not rely on cookies. It reads the conversation happening right now and determines whether an ad is relevant to that conversation. This is a fundamentally different targeting model, and it requires a fundamentally different approach to campaign planning.
The four types of contextual targeting
Not all contextual targeting is the same. Understanding the four types helps explain where ChatGPT’s approach fits and why it produces better results than older methods.
1. Keyword contextual targeting
The simplest form. An advertiser provides a list of keywords, and ads appear on pages or within content that contains those keywords. This is what Google Display Network and many programmatic platforms use as their baseline contextual option.
The limitation is precision. If you target the keyword “running,” your ad might appear in content about running shoes, running a business, running out of money, or running a marathon. The system matches on the word, not the meaning.
2. Category contextual targeting
A step up from keywords. Instead of individual words, advertisers select broad content categories like “Technology,” “Finance,” or “Health & Wellness.” The ad platform classifies content into these categories and serves ads accordingly.
Category targeting is scalable but imprecise. Selecting “Technology” means your ad appears alongside articles about smartphones, cybersecurity, AI research, vintage computing, and everything in between. The reach is broad, but the relevance is often low.
3. Semantic contextual targeting
This is where the technology gets interesting. Semantic targeting uses AI to analyze the actual meaning and intent of content, not just the words on the page. It understands context, nuance, and topic relationships.
A semantic system can distinguish between “running a marathon” and “running out of money.” It can tell the difference between a user researching CRM software to buy and a student writing an essay about CRM market trends. It can recognize that a conversation about “scaling customer support without hiring” is relevant to helpdesk software even though the word “helpdesk” never appears.
Semantic targeting delivers the highest accuracy of any contextual approach because it works at the meaning level, not the word level. Studies from GumGum and Integral Ad Science show that semantic contextual ads produce 2.5x higher brand recall and 40% stronger purchase intent compared to keyword-matched ads (GumGum/IAS Contextual Intelligence Study, 2024).
4. Sentiment contextual targeting
Sentiment targeting analyzes the emotional tone of content rather than just the topic. It determines whether content is positive, negative, neutral, or carries a specific emotional valence like frustration, excitement, or urgency.
This matters for brand safety. An ad for a luxury travel brand probably should not appear in a conversation where a user is expressing frustration about a ruined vacation. Sentiment targeting helps prevent these mismatches.
Where ChatGPT fits
ChatGPT’s ad system combines semantic contextual targeting with opt-in personalization. The primary signal is semantic: the system analyzes the full meaning and intent of the conversation thread. If users enable personalized ads in their settings, past chat history and ad interaction history add a secondary relevance layer.
This is the most advanced form of contextual targeting available on any major ad platform today. It operates at the meaning level, understands multi-turn conversation context, and does not depend on cookies, tracking pixels, or third-party data.
How to build topic clusters for ChatGPT
Since ChatGPT does not use keywords, your targeting strategy is built around topic clusters. A topic cluster is a group of related conversation prompts that share a common theme and map to a specific set of ad creatives and landing pages.
Building effective topic clusters is a five-step process.
Step 1: List your product categories
Start with the categories your product or service covers. If you sell project management software, your categories might include task management, team collaboration, resource planning, time tracking, and client reporting. If you sell running shoes, your categories might include road running, trail running, racing, recovery, and casual fitness.
Aim for three to eight top-level categories. Too few and you miss relevant conversations. Too many and you dilute your budget before you have data to optimize.
Step 2: Map 30–50 prompts per category
For each category, write 30 to 50 realistic prompts that your ideal customer would type into ChatGPT. These are not keywords. They are full conversational questions and problem descriptions written in natural language.
For a “task management” category, prompts might include:
- “How do I keep track of tasks across multiple projects without everything getting messy?”
- “My team uses Slack and Google Docs but we keep losing track of who’s doing what. What tools can help?”
- “I manage a team of 8 and need a simple way to assign and track tasks. We’re currently using spreadsheets.”
- “What’s the difference between Asana, Monday, and ClickUp for a small agency?”
Write these from the customer’s perspective, using their vocabulary. Talk to your sales team, support team, and customer success team. Read reviews of your product and competitors. Look at forums and communities where your customers ask questions. The goal is to capture the actual language your customers use when describing their problems.
Step 3: Group prompts by theme to form clusters
Review your prompts and group them by theme. Prompts about choosing between tools form one cluster. Prompts about managing remote teams form another. Prompts about integrating multiple tools form a third.
Each cluster should contain prompts that share a common underlying need. The cluster name should describe that need in plain language: “Choosing a project management tool,” “Managing tasks for remote teams,” “Integrating PM tools with existing stack.”
Step 4: Sort clusters by conversation stage
Not every prompt represents the same buying intent. Sort your clusters into three conversation stages:
- Research stage – the user is exploring a problem or category. They are learning, not buying. Example: “What are the benefits of using a project management tool?”
- Comparison stage – the user is evaluating specific options. They know what they want and are narrowing choices. Example: “How does Asana compare to Monday for a 10-person team?”
- Purchase-ready stage – the user is ready to act. They need pricing, trials, or a final push. Example: “Does ClickUp have a free plan for small teams?”
Stage mapping matters because each stage requires different ad creative and a different landing page experience.
Step 5: Match creative to each cluster and stage
Each cluster-stage combination gets its own set of ad creatives and a dedicated landing page. A research-stage prompt about remote team management should see a headline like “How teams manage tasks across time zones” and a landing page that educates. A purchase-ready prompt about the same topic should see “Free task tracking for remote teams” and a landing page with a trial signup.
This is where most advertisers fall short. They create one set of creatives and one landing page for all conversations. The advertisers who build creative mapped to specific clusters and stages will outperform those who take a one-size-fits-all approach.
Intent mapping worksheet
Use this worksheet to map each topic cluster to the right creative and landing page. Copy the table below and fill it in for each of your clusters.
| Field | Your entry |
|---|---|
| Cluster name | e.g., “Choosing a CRM for small teams” |
| Example prompt 1 | “What CRM works best for a 10-person sales team?” |
| Example prompt 2 | “I need a CRM that integrates with Gmail and costs under $30/seat” |
| Example prompt 3 | “We’re switching from Salesforce to something simpler. Options?” |
| Example prompt 4 | “Best CRM for an agency that manages 50+ client accounts” |
| Example prompt 5 | “How do I track leads and deals without a big CRM?” |
| Example prompt 6 | “CRM with built-in email sequences for outbound sales” |
| Example prompt 7 | “Is HubSpot free CRM actually free? What are the limits?” |
| Example prompt 8 | “I run a real estate team and need a CRM for tracking showings and follow-ups” |
| Example prompt 9 | “Comparing Pipedrive vs. Close vs. Copper for B2B startups” |
| Example prompt 10 | “What features should I look for in a CRM if I’m a solopreneur?” |
| Conversation stage | Research / Comparison / Purchase-ready |
| Matching headline (50 chars max) | e.g., “Free CRM built for teams under 20 people” |
| Matching description (100 chars max) | e.g., “Track leads, automate follow-ups, and close deals faster. No credit card required. Try free.” |
| Landing page URL | e.g., yoursite.com/crm-for-small-teams?utm_source=chatgpt |
Create one worksheet per cluster. If you have five clusters with three stages each, you will end up with 15 worksheets. This level of specificity is what separates high-performing ChatGPT ad campaigns from average ones.
Negative context exclusions
On Google Ads, negative keywords prevent your ads from showing when someone searches for a term that is related to your keywords but irrelevant to your product. ChatGPT has an equivalent concept: negative context exclusions. These exclude entire conversation topics where your ads would be irrelevant or wasteful.
Negative context exclusions are critical for budget efficiency. Without them, your ads may appear in conversations that generate impressions and even clicks but never convert because the user was never a potential customer.
Conversations to exclude
Educational queries from students. If you sell B2B software, students writing papers about your industry will trigger contextually relevant conversations. A student asking “explain how CRM software works for my business technology class” is not a buyer. Exclude academic and educational contexts unless your product is education-focused.
Career advice conversations. Users asking “how do I get a job in SaaS sales?” or “what skills do I need for a marketing career?” are discussing your industry but are not evaluating your product. These conversations burn budget without generating qualified traffic.
Competitor employee queries. People asking about your competitor’s internal processes, company culture, or hiring practices are likely employees or applicants, not potential customers. A conversation like “what is it like working at Salesforce?” is contextually close to CRM software but has zero purchase intent.
Academic research. Researchers studying market trends, technology adoption, or industry dynamics generate conversations that overlap with commercial topics but carry no purchase intent. A user writing a thesis on “adoption patterns of project management software in SMBs” is researching, not buying.
DIY and workaround seekers. Users asking “how do I build a CRM in Google Sheets?” or “free alternatives to project management software” are explicitly avoiding paid solutions. Showing them a paid product ad is unlikely to convert and may generate negative sentiment.
How to identify exclusion candidates
After your campaigns run for two to four weeks, review your performance data for patterns. Look for conversation topics that generate high impressions and clicks but near-zero conversions. These are your exclusion candidates. Add them to your negative context list and reallocate that budget toward clusters that convert.
Unlike Google’s negative keywords, which require exact or phrase matching, ChatGPT’s exclusions work at the semantic level. You describe the type of conversation to exclude, and the system recognizes variations of that context. This means fewer exclusion rules are needed to cover more ground.
Geographic and audience targeting
While ChatGPT ads are primarily contextual, there are geographic and audience constraints that affect campaign planning and addressable market sizing.
Current geographic availability
ChatGPT ads are currently available only in the United States. This means your ads will only be shown to users whose account settings indicate a US location. You can narrow targeting within the US by region, state, or metro area, but you cannot currently target internationally.
OpenAI has confirmed that Canada, Australia, and New Zealand are the next markets on the expansion roadmap, with availability expected during 2026 (The Information, March 2026). Broader international rollout is planned for later in 2026 and into 2027.
Audience eligibility
Not every US user is eligible to see ads. The eligible audience is defined by three criteria:
- Subscription tier: Free tier and Go plan ($8/month) users only. Plus ($20/month), Pro ($200/month), Business, Enterprise, and Education subscribers never see ads.
- Age: Users must be 18 or older.
- Login status: Users must be logged in to their ChatGPT account.
Estimating your addressable audience
ChatGPT has over 800 million weekly active users globally (OpenAI, March 2026). A significant share of that traffic comes from the United States, and the majority of US users are on the Free or Go tiers. However, only about 20% of eligible sessions currently show an ad (Financial Times, March 2026).
This means the practical addressable audience for any given campaign is a fraction of the total user base. But the 20% figure also means there is substantial room for OpenAI to increase ad load over time, which will expand the available inventory for advertisers.
For budget planning, assume a conservative reach estimate and let your actual impression data calibrate your projections. The key metric is not total reach but the number of contextually relevant conversations within your topic clusters that receive ad delivery.
Geographic narrowing within the US
If your business serves specific US regions, you can use geographic narrowing to focus your budget. A local HVAC company in Texas does not need nationwide reach. A SaaS company with US-only compliance certifications might want to exclude conversations from users who indicate international operations.
Geographic narrowing is particularly valuable for businesses with regional service areas, local retail footprints, or state-specific regulatory considerations. Combined with topic cluster targeting, it creates a highly focused audience of users who are both interested in your category and located in your service area.
Map intent to creative with Lapis
Building topic clusters and mapping them to creative is the most time-consuming part of ChatGPT ad campaign planning. Lapis accelerates this process by generating ad creatives mapped to your intent clusters and providing intelligence on what competitors are doing across platforms.
Creatives mapped to intent clusters
Instead of writing headlines and descriptions from scratch for each cluster-stage combination, describe your product and the conversation context in a text prompt. Lapis generates headline variations within the 50-character limit, descriptions within 100 characters, and images sized for ChatGPT’s thumbnail format. Each creative set is tailored to the specific cluster and buying stage you define.
If you have five clusters with three stages each, Lapis can generate the full creative matrix in a single session rather than the days of manual work it would otherwise require.
Competitor tracking for targeting intelligence
Understanding what your competitors are running on ChatGPT helps you identify gaps in their targeting and differentiate your messaging. Lapis tracks competitor ad creative across ChatGPT, Meta, Google, LinkedIn, and TikTok, giving you visibility into the headlines, descriptions, and visual approaches others in your category are using.
This intelligence helps you avoid running identical messaging to your competitors and identify topic clusters they may be underserving.
Forecasting per cluster
Lapis provides estimated impression ranges and budget projections for each topic cluster, helping you allocate spend where it will produce the best return. Start with your highest-confidence clusters, gather conversion data, and then expand to adjacent clusters as your performance data grows.
Try Lapis for free to start mapping your intent clusters to ChatGPT ad creatives.
For a comprehensive overview of the platform, read our complete guide to ChatGPT ads. For step-by-step creative instructions, see how to create ads for ChatGPT. For headline optimization strategies, check our ChatGPT ads headline testing guide.
