Prompts are the new keywords
For twenty years, paid search rewarded advertisers who could guess the two-to-five-word fragment a buyer would type into a search box, then bid on it. That skill is becoming less useful every quarter. When people talk to ChatGPT, they do not type fragments. They describe a situation in full sentences: what they are trying to do, what they have already tried, their budget, their constraints, and the decision they are weighing. The prompt is a paragraph of declared intent, and that paragraph, not a keyword, is what ChatGPT’s ad system reads to decide whether your ad belongs in the answer.
This is why the shift matters for advertisers. On Google you buy your way onto a keyword. On ChatGPT you earn a placement by matching the meaning of what a buyer is asking, so the work moves upstream: you have to actually know the questions your customers ask before they buy. The advertisers who win are the ones who have cataloged those prompts and mapped each one to a relevant, helpful response.
25%
Projected decline in traditional search volume by 2026 as buyers shift to AI assistants
The demand is already moving. In a PYMNTS Intelligence study, the share of shoppers who used ChatGPT to research a purchase jumped from roughly 2% to about 30% in two years, and nearly half of online shoppers now enlist AI somewhere in their most recent purchase journey (PYMNTS Intelligence, 2026). Gartner’s 25% figure is a projection, and search has evolved rather than collapsed, but the direction is not in dispute: a large and growing share of buying questions is now being asked inside a chat window.
30%
Share of shoppers using ChatGPT for product research, up from about 2% two years earlier
ChatGPT ads reflect this reality. A single Sponsored card appears below the answer, matched to the conversation by context hints and topic clusters rather than keyword bids. The mechanics of building those hints are covered in our ChatGPT ads targeting guide. This article is the demand side: the catalog of real prompts buyers type, and how to read the intent behind them.
The intent funnel: from curiosity to checkout
Not every prompt carries the same buying intent. The same person asks very different questions at the start of a decision than they do the moment before they act. Sorting prompts into four stages, awareness, consideration, decision, and purchase, tells you which ad response fits and how hard to push.
| Stage | What the buyer is doing | Example prompt | Ad response that fits |
|---|---|---|---|
| Awareness | Naming a problem, not a product | “why do my Instagram ads stop working after a week?” | Educational card that frames the problem and your category |
| Consideration | Comparing approaches or categories | “what’s the best way to make ad creative without a designer?” | Category positioning with one concrete differentiator |
| Decision | Comparing named options | “AdCreative vs Lapis for a small marketing team” | Head-to-head proof: pricing, ratings, what sets you apart |
| Purchase | Ready to act on price or a trial | “is there a free AI ad generator I can try today?” | Low-friction offer: free tier, no card, start in minutes |
Stage matters because it changes both the message and the destination. An awareness prompt wants education, so a hard trial offer feels premature and converts poorly. A purchase prompt wants friction removed, so an educational explainer wastes the moment. The same product needs different creative and different landing pages for each stage.
43%
Of US online shoppers used an AI assistant to research a purchase in the past 90 days
One nuance shapes decision and purchase-stage creative: buyers still verify. In the same Product.ai research, 86% of people who used AI for product research checked the recommendation against another source before buying. That means your decision-stage ads should carry proof, ratings, transparent pricing, and specifics, because the buyer is going to fact-check you anyway. Give them the evidence up front.
A prompt taxonomy by industry
The prompts your buyers type are specific to your category. Below is a working taxonomy across six industries, each with real example prompts mapped to stage and the ad angle that fits. Use these as a starting template, then replace the examples with the exact language your own customers use.
SaaS and software
Software buyers describe a workflow that is breaking and the constraints on any fix, such as team size, budget per seat, and the tools they already run.
| What the buyer types | Stage | Ad angle that fits |
|---|---|---|
| “how do I keep projects from slipping without micromanaging?” | Awareness | Frame the pain, then introduce the category |
| “Asana vs Monday vs ClickUp for a 12-person agency” | Decision | Comparison plus your differentiator and a use-case fit |
| “project management tool under $10 per user with time tracking” | Purchase | Price-anchored offer with the exact feature named |
| “alternative to Jira that non-engineers will actually use” | Decision | Position as the approachable switch with easy migration |
E-commerce and DTC
Physical-product prompts are the most likely to surface shopping units, and they get very specific about attributes, budget, and use case.
| What the buyer types | Stage | Ad angle that fits |
|---|---|---|
| “best moisturizer for oily, acne-prone skin under $30” | Purchase | Product spotlight with price and the specific attribute |
| “what should I look for in a standing desk?” | Awareness | Buying-guide card that educates, then links a collection |
| “Bombas vs Darn Tough for hiking socks” | Decision | Head-to-head with reviews and durability proof |
| “lightweight stroller that folds for air travel” | Consideration | Use-case fit, ships-fast, direct product link |
Local and service businesses
Local prompts carry urgency and geography. The words “near me” and “open now” are among the highest-intent phrases a buyer can type.
| What the buyer types | Stage | Ad angle that fits |
|---|---|---|
| “emergency plumber near me open now” | Purchase | Local, available now, call-or-book CTA |
| “how much does it cost to rewire a house?” | Consideration | Price transparency plus a free-quote offer |
| “best HVAC company in Austin with financing” | Decision | Local proof, reviews, financing highlighted |
| “do I need a permit to remodel my kitchen?” | Awareness | Educate, then capture with a consultation offer |
Finance and fintech
Money prompts revolve around rates, fees, and whether a product is worth it. Trust and specifics matter more here than anywhere else.
| What the buyer types | Stage | Ad angle that fits |
|---|---|---|
| “best high-yield savings account for an emergency fund” | Purchase | Rate-led card, FDIC insured, open in minutes |
| “how do I lower my business’s payment processing fees?” | Consideration | Savings-led angle with transparent pricing |
| “Mercury vs Brex for an early-stage startup” | Decision | Comparison, perks, and a no-fee structure |
| “is a Roth IRA worth it if I’m self-employed?” | Awareness | Educate, then offer a low-friction account open |
B2B
B2B prompts embed the buying committee’s constraints: existing stack, compliance needs, headcount, and the pressure to prove ROI.
| What the buyer types | Stage | Ad angle that fits |
|---|---|---|
| “how do we cut SaaS spend without losing tools the team needs?” | Consideration | ROI framing with an audit or savings offer |
| “best CDP for a 200-person company already on Snowflake” | Decision | Fit plus integration proof and a demo CTA |
| “vendor that handles SOC 2 evidence collection automatically” | Purchase | Name the exact capability, book-a-demo |
| “alternative to Salesforce for a services firm that hates complexity” | Decision | Simpler-switch positioning with migration help |
Education and info products
Learners ask about outcomes and whether a program is worth the time and money. Job results and guarantees do the persuading.
| What the buyer types | Stage | Ad angle that fits |
|---|---|---|
| “best way to learn data analysis if I’m switching careers” | Consideration | Outcome-led angle with curriculum fit |
| “is a PMP certification worth it in 2026?” | Awareness | Educate on the cert’s ROI, then an enroll CTA |
| “affordable online course for the AWS solutions architect exam” | Purchase | Price plus pass-rate proof, enroll now |
| “Coursera vs Udemy for a beginner UX bootcamp” | Decision | Comparison with job outcomes and a guarantee |
Reading intent signals in a prompt
You do not need a model to tell a researcher from a buyer. The words do it for you. Certain phrase patterns reliably separate someone learning about a category from someone about to spend money, and one of the clearest signals on ChatGPT is transactional language itself.
9x
Higher share of transactional-intent queries on ChatGPT than on traditional search
Profound’s analysis of more than 50 million ChatGPT prompts found transactional intent at 6.1% of queries versus 0.6% on traditional search, roughly a ninefold jump, with commercial intent close behind. In other words, a large slice of ChatGPT traffic is people actively trying to buy or shortlist, not just browse. The phrases below tell you which slice you are looking at.
| Phrase pattern | What it signals | Intent | Example |
|---|---|---|---|
| “best X for Y” | Active evaluation for a specific use case | High | “best CRM for a real estate team” |
| “X vs Y” | Down to a shortlist | High | “Pipedrive vs HubSpot” |
| “under $N” / “cheapest” | Budget set, ready to transact | High | “email tool under $50 a month” |
| “alternative to X” | Unhappy with an incumbent, switching | High | “alternative to Mailchimp” |
| “near me” / “open now” | Local, immediate need | High | “dentist near me open Saturday” |
| “how much does X cost” | Price-checking before commitment | Medium-high | “how much does bookkeeping cost” |
| “is X worth it” | Final rationalization before buying | Medium-high | “is QuickBooks worth it for a freelancer” |
| “what is X” / “how does X work” | Learning the category | Low | “what is a CDP” |
| “free” / “DIY” / “template” | Explicitly avoiding paid solutions | Low | “free CRM template in Google Sheets” |
Mapping prompt clusters to context hints
A prompt cluster is a group of prompts that share one underlying need. A context hint is the plain-language description you give ChatGPT’s ad system so it recognizes that cluster in a live conversation. Turning one into the other is the bridge between this demand-side catalog and an actual campaign. The full mechanics live in our targeting guide; here is the end-to-end move.
Start with a cluster. Say you sell a lightweight CRM and you have grouped these prompts:
- “what CRM works best for a 10-person sales team?”
- “alternative to Salesforce that is easier to set up”
- “CRM under $30 per seat that connects to Gmail”
- “we’re outgrowing spreadsheets for tracking deals, what should we use?”
Name the cluster in the buyer’s language: “Choosing a simple CRM for small sales teams.” Then write the context hint as a description of the conversation, not a keyword: “Conversations where a small business or sales team is evaluating or switching CRM software, focused on ease of setup, Gmail integration, and per-seat cost.” Add exclusions for the contexts that look similar but never buy, such as students, job seekers, and people asking how to build a CRM in a spreadsheet.
Finally, match the creative to the cluster and its dominant stage. Because this cluster is comparison and purchase heavy, the ad leads with proof and price inside ChatGPT’s format: a 50-character headline, a 100-character description, and a 256×256 image, shown as one Sponsored card. A headline like “Free CRM built for teams under 20 people” with “Connects to Gmail, sets up in minutes, no card required” answers the exact prompt behind the conversation.
Build your own prompt map
The taxonomy above is a template. Your real advantage comes from the prompts unique to your customers, in their own words. The exercise is straightforward: write 30 to 50 real prompts, group them by stage, and prioritize. Do not invent them at a whiteboard. Pull them from where your customers already talk.
- Read the last 50 sales-call notes and support tickets. Copy the exact phrasing customers use to describe their problem.
- Mine reviews of your product and your competitors, plus the questions in relevant subreddits and communities.
- Check your search console and site search for the queries that already bring people in.
- Ask ChatGPT itself, in the voice of your customer, what it would ask before buying your category, then sanity-check the output against the real language above.
Once you have your list, tag each prompt with a stage and an intent level using the signal patterns from the previous section. Then group the tagged prompts into clusters of shared need. Aim for three to eight clusters to start; more than that and you will spread a first budget too thin to learn anything.
Prioritizing prompt clusters by value
Not every cluster deserves the same budget. Score each one on three axes, then fund the top of the list first. A simple additive model works: rate intent, estimated volume, and margin from 1 to 5, and add them for a score out of 15. Intent is how ready the cluster is to buy, volume is how often those prompts occur, and margin is how much you make when they convert.
| Prompt cluster | Intent | Volume | Margin | Score | Fund first? |
|---|---|---|---|---|---|
| “best [product] for [niche]” | 4 | 5 | 4 | 13 | Yes |
| “alternative to [incumbent]” | 5 | 3 | 4 | 12 | Yes |
| “[you] vs [competitor]” | 5 | 2 | 5 | 12 | Yes |
| “how does [category] work” | 2 | 5 | 3 | 10 | Later |
| “free [category] template” | 1 | 4 | 1 | 6 | No |
The scoring is deliberately simple so you can build it in a spreadsheet in an afternoon and defend every number to your team. High-intent, high-margin clusters earn budget first even when volume is modest, because they convert and pay for the learning. Low-intent, low-margin clusters like “free template” seekers are usually better left to organic content than paid placement.
From prompts to creative with Lapis
A prompt map is only useful once it becomes campaigns. Lapis is built for exactly this handoff: it turns your understanding of buyer prompts into ready-to-run creative and helps you decide which clusters to fund.
Start with AI Audience personas. Instead of guessing, you build a persona for each segment and Lapis reveals how that segment actually phrases its questions, giving you a head start on the prompt map and surfacing clusters you might have missed. Then Performance Forecasting estimates the return on each cluster so you can prioritize the prompt groups worth funding before you spend, rather than after. That matters because ChatGPT inventory is not cheap: CPMs run about $25 to $60 and CPCs about $3 to $5, with self-serve access on ads.openai.com carrying a roughly $5,000 monthly minimum as of April 2026.
1.5–4x
Higher conversion rate for ChatGPT ad traffic compared with equivalent Google Search campaigns
Once you know which clusters to back, generate the creative. From a single prompt, Lapis produces ChatGPT ads alongside Meta, Google, Reddit, and LinkedIn variations in under three minutes, so each cluster and stage gets matched creative instead of one generic ad stretched across every conversation. That is what turns the higher conversion rates ChatGPT delivers into real return: the right message meeting the right prompt.
Lapis is a Y Combinator F25 company with a 5.0 rating on G2 and more than 10,000 campaigns created on the platform. You can start on the free tier; the Pro plan at $599 per month is the recommended tier for teams running ChatGPT and multi-platform campaigns at volume. Start with Lapis and build the creative for your highest-value prompt clusters today.
Get started
The advertisers who win on ChatGPT are the ones who understand the demand side first: the real prompts their buyers type, the intent behind the language, and which clusters are worth funding. Build the prompt map, read the signals, score your clusters, then let the creative and forecasting fall out of that work.
Start with Lapis to turn your prompt map into ChatGPT-ready creative. For the targeting mechanics behind context hints and topic clusters, read the ChatGPT ads targeting guide, and for tuning clusters toward conversions see the conversion targeting guide. To build the ad units themselves, see how to create ads for ChatGPT. To weigh paid placement against earned mentions, read paid ads vs. organic mentions, and for the full picture start with the complete guide to ChatGPT ads.
