Marketing AI Products Playbook: How to Position, Prove, Price and Sell an AI Product

By Saroj Jha, AAJ · Published October 1, 2026 · Pairs with the Positioning Statement Generator and the Five-Second Message Test

Sell an AI product on the job it does and the result it delivers, not on the model behind it. Then prove it on the buyer’s own work, and never claim more than you have tested. The eight steps below show how, with one example company from start to finish.

Every software company now says it has AI. That makes “AI-powered” close to meaningless as a message, and it raises the bar for proof. Many buyers have heard the promise before, and some have tried AI without seeing results.

In this playbook I walk through what is different about marketing an AI product, and the order in which I would work through positioning, proof, claims, price and sales with a founder. One company runs through every step.

Figure 1 — The eight steps of this playbook

Figure 1 shows the eight steps in order: message, differentiate, proof kit, pilot, claims, price, channels and after the sale.

Why is marketing an AI product different?

Three facts about today’s buyers shape everything else.

Use is high, but value is rare. In McKinsey’s 2026 global survey of 1,719 respondents, nearly nine in ten organizations use AI regularly in at least one function, but only 37% attribute any profit (EBIT) impact to it, and only about 6% qualify as high performers (McKinsey, 2026). BCG’s 2025 survey of 1,250 executives found 60% of companies report minimal revenue and cost gains from AI (BCG, 2025). Many buyers may already have tried AI without seeing results.

Trust is low. In the University of Melbourne and KPMG’s 2025 study of 48,340 people in 47 countries, 66% use AI regularly but only 46% are willing to trust AI systems (Gillespie and others, 2025).

More people join the decision. Forrester’s 2026 buyer research, based on nearly 18,000 business buyers, found the buying group doubles, from seven to 14 people, when a purchase includes generative AI features (Forrester, 2026). Expect security, legal and data teams at the table.

My response to all three is the same: be specific, prove value before asking for commitment, and answer the risk questions before they are asked.

Step 1: Lead with the job and the result, not “AI”

In my view “AI” rarely belongs in the headline. Lead with who the product is for, the job it does and the result. Explain the AI when you explain how it works.

There is some evidence for this. In a peer-reviewed set of experiments with more than 1,000 US adults, mentioning “artificial intelligence” in a product description lowered emotional trust and purchase intention, and the effect was stronger for higher-risk products (Cicek, Gursoy and Lu, 2025). That study is about consumers, not business buyers, and I have found no equivalent B2B study. But it fits the public mood: in Pew Research Center’s 2025 survey of 5,023 US adults, 50% were more concerned than excited about AI in daily life, up from 37% in 2021 (Pew Research Center, 2025).

Write the message in five parts, in this order: who it is for, the job, the result, how the AI does it, and the limit. The limit is not a weakness. It answers the first question a careful buyer will ask.

Figure 2 — The five-part message order, with Example Co.’s wording (fictional)

Figure 2 shows the five-part message order with Example Co.’s fictional wording: who it is for, the job and the result go in the headline; how the AI does it and the limit go in the body copy.

Do this: write your headline in the five-part order, then test it with the Five-Second Message Test. If “AI” appears in the first three parts, move it down.

Step 2: Compete on what a model cannot copy

Do not compete on the model. Model performance is converging and access is getting cheap. Stanford’s AI Index found the cost of reaching GPT-3.5-level performance fell from $20 to $0.07 per million tokens between November 2022 and October 2024, a more than 280-fold drop (Stanford HAI, 2025). As of March 2026, the top models from four companies were within 25 points of each other on the Arena leaderboard (Stanford HAI, 2026). Whatever model advantage you have today, a competitor can likely buy soon.

You also have a new competitor: the buyer’s own team. In McKinsey’s 2026 survey, 32% of respondents said their organization had decided against buying one or more software products or features because they could be built internally with agentic coding tools (McKinsey, 2026). Name “we could build this ourselves” as a competitive alternative, and for many products “we could use a free chatbot” too.

Differentiate on what is hard to copy:

Source of advantageWhat it looks like in your marketing
Workflow fitBuilt into the tools and steps the buyer already uses
Data and contextTrained or tuned on domain data the buyer cannot easily assemble
Measured accuracy on the taskPublished evaluation on realistic cases, with known limits
Trust and governanceClear answers on data use, security, human review and audit
Outcome ownershipService, onboarding and support that deliver the result

What Duolingo’s and Chegg’s filings show

Two education companies’ public filings show both sides of this shift. These are what the documents say, not an account of how either company made its choices internally.

Duolingo sold AI as a paid upgrade and located its advantage around the model. In its first-quarter 2023 shareholder letter, Duolingo said it had “introduced two AI features as part of our new subscription tier, Duolingo Max, that are built with GPT-4: Explain My Answer and Roleplay.” It described them by what they are for: “Both of these features are aimed at teaching better by emulating a human tutor experience.” And it placed its edge outside the model: “our competitive advantage comes from our vast user data, in-house machine learning and AI expertise, gamified approach, and brand. We’re not using GPT-4 ‘off the shelf.’” (Duolingo Q1 2023 shareholder letter). Its annual report for 2023 describes Max as “a premium subscription tier launched in 2023 and offered to a portion of our user base and priced higher than Super Duolingo” (Duolingo 10-K for 2023).

Chegg met AI first as a competitor. Chegg’s annual report for 2023 says: “New technologies, including those based on AI, can provide students with more immediate responses than traditional tools. Over time, the accuracy of these tools and their ability to handle complex questions may improve, which may be disruptive to education technology businesses, such as ours.” The same report states: “In 2023, 7.7 million students subscribed to our Subscription Services, a decrease of 6% year over year from 8.1 million in 2022.” Its response was to build AI into its own product: “Beginning in September 2023, we started to roll out the first phase of our new AI-powered user experience, and we are continuing to make significant investments in AI initiatives.” (Chegg 10-K for 2023). I place these facts side by side; they do not show how much of the decline came from AI.

Figure 3 — Duolingo and Chegg on AI, verbatim from their SEC filings

Figure 3 sets the Duolingo and Chegg quotes from their SEC filings side by side, word for word as quoted above.

The lesson I take for a startup: free, general AI tools can become the alternative your buyer compares you with, and the durable part of your position is what you build around the model.

Do this: write down “build it ourselves” and “use a free chatbot” as named alternatives, with one sentence on why a buyer should choose you over each. The competitive analysis post covers mapping all your alternatives, including the status quo.

Step 3: Build a proof kit

Buyers want proof that it works on their work, with their data. Accuracy is the risk companies work on most. In McKinsey’s 2026 survey, inaccuracy was the AI risk respondents most often said their organizations are working to mitigate, cited by just over half (McKinsey, 2026). In its 2025 survey, 51% of organizations using AI had seen at least one negative consequence, and nearly a third reported consequences from inaccuracy (McKinsey, 2025).

Buyers also still want a person to check the answer. In Gartner’s survey of 645 B2B buyers, 69% prefer to validate AI-generated insights with sales reps (Gartner, 2026).

A useful proof kit has five parts:

  1. A test on the buyer’s data, with success measures agreed before it starts.
  2. A before-and-after metric: time, cost, error rate or revenue on one workflow.
  3. An honest limits page: what the AI gets wrong, and how errors are caught.
  4. Security and data-use answers: where data goes, whether it trains models, who can see it.
  5. A human in the loop: who reviews outputs, and how the buyer stays in control.
Figure 4 — The proof kit, with Example Co.’s version of each item (fictional)

Figure 4 lists the five parts of the proof kit with Example Co.’s fictional version of each.

Do this: write the limits page first. If you cannot say where the product fails, you are not ready to publish accuracy numbers. The Win-Loss Analyzer helps you learn which parts of the kit decided past deals.

Step 4: Run pilots that end in a decision

Trials are common, but they are not sales. Forrester found more than 60% of business buyers purchase some form of trial, but just over a third of them planned to convert to a full paid version with the same provider (Forrester, 2026). Among buyers making purchases of $10 million or more, 78% trial first (Forrester, 2026).

So treat a pilot as a test with an end date. Before it starts, agree on one workflow and one success measure, the data the test will use and how it is protected, and a decision date. At the end, walk the buyer through the results in person; that is where the rep’s validation role from Step 3 pays off.

Figure 5 — Example Co.’s 30-day pilot with a decision date (fictional)

Figure 5 shows Example Co.’s fictional 30-day pilot: before day 1, weeks 1 to 4, the day 30 review and the decision to buy, extend once with a new date, or stop.

Do this: never start a pilot without a decision date in writing.

Step 5: Check every claim before you publish

Claim only what you can prove when you say it. Regulators now treat exaggerated AI claims as ordinary deception. This is orientation, not legal advice.

  • The FTC. Launching a 2024 crackdown on deceptive AI claims, the FTC said “there is no AI exemption from the laws on the books” (FTC, 2024). In 2025 it acted against Workado, which marketed an AI-content detector as 98% accurate; the FTC’s complaint said independent testing found 53% accuracy on general-purpose content (FTC, April 2025). The final order requires “competent and reliable evidence” for such claims (FTC, August 2025).
  • The SEC. It has charged firms with “AI washing”: claiming AI capabilities they did not have (SEC, 2024). In 2025 it charged a startup founder who raised over $42 million while claiming his app completed purchases with AI; the SEC alleged the app “relied in large part on contract employees to manually input orders” (SEC, 2025). Investor materials count as marketing too.
  • “Agent washing.” Gartner estimates only about 130 of the thousands of vendors claiming agentic AI offer genuine agents, and predicts over 40% of agentic AI projects will be canceled by the end of 2027 (Gartner, 2025). Call something an agent only if it acts on its own.
  • The EU AI Act. From August 2, 2026, providers must tell people when they are interacting with an AI system unless it is obvious, and providers of generative AI systems must mark AI-generated content in a machine-readable way; systems already on the market have until December 2, 2026 for the marking rule (European Commission, 2026). If you sell into the EU, check these duties.
  • Children’s data. If your product serves children, check children’s privacy law as well. Under the US rule that implements the Children’s Online Privacy Protection Act (COPPA), a “child” is an individual under 13, and personal information includes “A photograph, video, or audio file where such file contains a child’s image or voice” (16 CFR 312.2).

A simple rule for every claim: a number needs a test behind it, dated and repeatable; “autonomous” needs no human doing the work; “secure” needs a named standard or audit.

Figure 6 — The claim check, from claim to question to fix

Figure 6 shows four claim checks, each running from the claim to a question to the fix if the answer is no: a number, “autonomous” or “agent”, “secure” or “private”, and children’s voices.

Do this: list every claim on your site and in your deck, and next to each write the evidence and its date. Delete what has none.

Step 6: Price on a metric that grows with value

Pick a price metric that grows with the value the buyer gets and covers your own running costs. Practice is still settling. Bain’s 2026 review of about 200 B2B software companies’ pricing found about four in five vendors introducing AI pricing chose capacity models, such as prepaid credits or entitlements. Among companies introducing a hybrid AI meter, about 55% meter output, about 35% meter effort and only about 10% meter outcomes. About one in five AI-native companies still priced mostly per seat (Bain, 2026).

Buyers want flexibility. In a McKinsey survey of 150 enterprise software buyers (October 2024), 65% said being able to move usage or spending commitments from one product to another was very or extremely important (McKinsey, 2025).

No study yet shows which model performs best, so choose by fit:

ModelWorks whenWatch out for
Per seatValue tracks the number of usersAI reduces seats, and revenue with them
Usage or creditsValue and cost both rise with useBuyers fear unpredictable bills
Output-basedA clear unit of work (a document, a call, a case)Defining the unit in a way both sides accept
Outcome-basedThe result is measurable and attributableDisputes over what caused the result

A hybrid, a base fee plus included usage, is a common compromise.

Figure 7 — Example Co.’s base-plus-usage price at three center sizes (fictional)

Figure 7 shows Example Co.’s fictional price: a $200 monthly base fee that includes up to 40 active readers, plus $4 a month for each extra reader, so a large center with 60 readers pays $280.

Do this: model your price at three customer sizes with the Pricing ARPU Modeler, and check your running costs at each. The SaaS pricing models post covers the trade-offs in more depth.

Step 7: Go where buyers can see it work

Choose channels where buyers can see the product work. Three matter most:

  • Hands-on trials and product-led growth. Let people try it on a real task quickly. The Product-Led Growth Playbook covers this.
  • AI answers. Forrester found 94% of business buyers use AI during their buying process (Forrester, 2026). If AI assistants do not mention your product, you are missing a growing research channel. The GEO and AEO Playbook covers being cited.
  • Expert sales. With 14-person buying groups and buyers checking AI answers with reps, you need people who can run a proof and answer security questions.

Content that shows your evaluation method, your known limits and real before-and-after examples builds more trust than feature lists. Use fictional or permission-based examples only.

Do this: ask the three AI assistants your buyers use most to describe your product, and fix your public pages until the answers are right.

Step 8: Keep proving value after the sale

The real test starts after the contract. In McKinsey’s 2026 survey, 44% of respondents said AI is scaling across their enterprise, up from 38% a year earlier, so most organizations have yet to scale it (McKinsey, 2026). Gartner says at least half of generative AI projects were abandoned after proof of concept by the end of 2025, though it does not publish a sample for that figure (Gartner, 2026). If the customer’s project stalls, your renewal goes with it.

Plan onboarding that reaches the first measurable result fast, report value back to the buyer every quarter, and train the users, not just the admins. The Customer Onboarding and Retention Playbook covers the method.

Figure 8 — Example Co.’s plan from day 1 to renewal (fictional)

Figure 8 shows Example Co.’s fictional plan from day 1 to renewal: day 1, week 1, month 1, each quarter and renewal.

Do this: define the first result a new customer should see, and the date by which they should see it. Then check every new account against that date.

What changes by stage?

Pre-seed. Pick one job for one buyer. Measure accuracy on real cases and write down the limits. Keep claims narrow.

Seed. Build the proof kit and a repeatable pilot with agreed success measures. Test a hybrid price with a few customers. Publish your security and data-use page.

Series A–B. Add security and legal content for bigger buying groups, a customer value report, and expansion pricing. Review every public claim against current evidence each quarter.

What mistakes should I avoid?

  • Leading with the model. Buyers buy outcomes, and the model is replaceable.
  • Claiming accuracy you have not tested. Regulators now act on it.
  • Calling a workflow an agent. Gartner calls this “agent washing”.
  • Open-ended free pilots. Just over a third of trial buyers plan to convert; agree measures and a decision date up front.
  • Ignoring “build it ourselves” and free chatbots. Both are now common alternatives.
  • Pricing that punishes success. Per-seat pricing can shrink as your AI replaces work.

Use this on AAJ

Frequently Asked Questions

How do I market an AI product?

Lead with the job it does and the result it delivers, prove it on the buyer’s own data, state its limits and data use clearly, and price on a metric that tracks value.

Should I say “AI-powered” in my messaging?

Not as the headline. Buyers respond to outcomes, and a consumer study found that mentioning AI can lower trust. Explain the AI where you explain how the product works.

How do I differentiate an AI product when competitors use the same models?

Compete on workflow fit, domain data, measured accuracy on the task, trust and governance, and the service that delivers results. Duolingo’s 2023 shareholder letter, for example, located its advantage in its user data, AI expertise, gamified approach and brand, not in the model.

What AI marketing claims get companies in trouble?

Accuracy numbers without testing, claims that AI does work people actually do, and “agent” labels on simple automation. The FTC and SEC have both taken action on such claims.

How should I price an AI product?

About four in five vendors introducing AI pricing now use capacity models, such as prepaid credits, often alongside a base fee. Choose the metric that rises with the buyer’s value and covers your running costs.

Sources

  1. McKinsey & Company (2026). “The state of AI in 2026: On the road to ROI.” Survey of 1,719 respondents, May–June 2026. mckinsey.com
  2. McKinsey & Company (2025). “The state of AI in 2025: Agents, innovation, and transformation.” Survey of 1,993 participants. mckinsey.com
  3. BCG (2025). “The Widening AI Value Gap: Build for the Future 2025.” Survey of 1,250 executives. bcg.com
  4. Gillespie, N., Lockey, S., Ward, T., Macdade, A. and Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne and KPMG. doi:10.26188/28822919. figshare.unimelb.edu.au
  5. Forrester (2026). “The State Of Business Buying: Risk-Averse Buyers Demand Proof, Not Promises.” Blog; nearly 18,000 buyers. forrester.com
  6. Forrester (2026). “Forrester: 2026 The State of Business Buying.” Press release, January 21, 2026. forrester.com
  7. Cicek, M., Gursoy, D. and Lu, L. (2025). “Adverse impacts of revealing the presence of ‘Artificial Intelligence (AI)’ technology in product and service descriptions on purchase intentions.” Journal of Hospitality Marketing & Management, 34(1); first published online 2024. doi:10.1080/19368623.2024.2368040
  8. Pew Research Center (2025). “How Americans View AI and Its Impact on People and Society.” Survey of 5,023 US adults, June 2025. pewresearch.org
  9. Stanford HAI (2025). Artificial Intelligence Index Report 2025. hai.stanford.edu
  10. Stanford HAI (2026). The 2026 AI Index Report, Chapter 2: Technical Performance. hai.stanford.edu
  11. Duolingo, Inc. (2023). Q1 2023 shareholder letter, May 2023, filed with the US Securities and Exchange Commission. sec.gov
  12. Duolingo, Inc. (2024). Annual report on Form 10-K for the fiscal year 2023, filed February 29, 2024. sec.gov
  13. Chegg, Inc. (2024). Annual report on Form 10-K for the fiscal year 2023. sec.gov
  14. Gartner (2026). “Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights.” Press release, May 20, 2026; 645 buyers. gartner.com
  15. Federal Trade Commission (2024). “FTC Announces Crackdown on Deceptive AI Claims and Schemes.” ftc.gov
  16. Federal Trade Commission (2025). “FTC Order Requires Workado to Back Up Artificial Intelligence Detection Claims.” April 28, 2025. ftc.gov
  17. Federal Trade Commission (2025). “FTC Approves Final Order against Workado.” August 2025. ftc.gov
  18. US Securities and Exchange Commission (2024). Press Release 2024-36, “AI washing” charges. sec.gov
  19. US Securities and Exchange Commission (2025). Litigation Release 26282. sec.gov
  20. Gartner (2025). “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027.” Press release, June 25, 2025. gartner.com
  21. European Commission (2026). “Transparency obligations under Article 50 of the AI Act” (FAQ). digital-strategy.ec.europa.eu
  22. Code of Federal Regulations, 16 CFR 312.2, Children’s Online Privacy Protection Rule, definitions. ecfr.gov
  23. Maltiel, E. and Sandberg, J. (2026). “AI Pricing: A Reality Check on Effort, Usage, and Outcomes.” Bain & Company. bain.com
  24. Khanna, Mittal, Wu and Castillo (2025). “Upgrading software business models to thrive in the AI era.” McKinsey & Company. mckinsey.com
  25. Chandrasekaran, A. (2026). “Why 50% of GenAI Projects Fail — And How to Beat the Odds.” Gartner. gartner.com

More in Industry Playbooks

Part of the Industry Playbooks hub - see all 7 resources on this topic.

Also useful in Strategy & Positioning, Pricing & Monetization.