How to Market an AI Product (When Every Product Says It Has AI)

  1. Blog
  2. Strategy & Positioning

By Saroj Jha · October 1, 2026 · 9 min read

Market an AI product by what it does for a specific buyer, not by the fact that it uses AI. Then show it working on the buyer’s own work, and never claim accuracy or autonomy you have not tested.

In three numbers

  • 52% of US adults were more concerned than excited about AI in 2026, up from 37% in 2021[1]
  • 280-fold fall in the price of GPT-3.5-level performance, November 2022 to October 2024[2]
  • 53% accuracy found in testing for an AI detector marketed as 98% accurate, per the FTC[3]

“AI-powered” used to set a product apart. Now AI is everywhere: in McKinsey’s 2026 survey of 1,719 people in 97 countries, nearly nine in ten said their organizations use AI regularly in at least one business function[4]. When every vendor can say the same word, the word stops working.

It also no longer reassures people. In Pew Research Center’s June 2026 survey of 3,488 US adults, 52% said they were more concerned than excited about the increased use of AI in daily life, up from 37% in 2021[1]. In experiments with more than 1,000 US adults, simply mentioning “artificial intelligence” in a product description lowered emotional trust and purchase intention[5]. Business buyers may react differently, but none of this suggests the word sells on its own. Here is what I would do instead, in six steps.

Figure 1 Since 2023, about half of US adults have been more concerned than excited about AI; in 2021 it was 37%

How the increased use of AI in daily life makes US adults feel

  • More concerned than excited
  • Equally concerned and excited
  • More excited than concerned
  • 2021 37% 45% 18% Equally 45%, more excited 18%
  • 2022 38% 46% 15% Equally 46%, more excited 15%
  • 2023 52% 36% 10% Equally 36%, more excited 10%
  • 2024 51% 38% 11% Equally 38%, more excited 11%
  • 2025 50% 38% 10% Equally 38%, more excited 10%
  • 2026 52% 37% 9% Equally 37%, more excited 9%
Show the numbers as a table
Year More concerned Equally More excited
2021 37% 45% 18%
2022 38% 46% 15%
2023 52% 36% 10%
2024 51% 38% 11%
2025 50% 38% 10%
2026 52% 37% 9%

Pew Research Center, short read of August 18, 2026[1]: survey of 3,488 US adults, June 22 to 28, 2026, and the same question in Pew’s surveys from 2021 to 2025. Where a year adds to less than 100%, the rest gave no answer. Bars drawn to scale.

Running case: Example Co. Fictional

Example Co. is a seed-stage EdTech startup with an AI reading coach for children in kindergarten to grade 3. A child reads aloud, the coach listens and flags words to practice, and adults get a weekly progress report. It sells first to independent tutoring centers, for $200 a month per center. Its first homepage said “AI-powered reading assessment platform”. This post follows the team as it fixes that. Example Co. and all its numbers are illustrative.

Figure 2 Each of the six steps answers a question a careful buyer asks before buying an AI product
The buyer asks The step that answers it

  1. 1 “What does it do for someone like me?”

    Step 1: Say the job first Who it is for, the job and the result; then how the AI does it, and its limit

  2. 2 “Why not build it, or use a free chatbot?”

    Step 2: Compete on what the model cannot give your rivals Workflow fit, domain data, tested accuracy and support

  3. 3 “Will it work on my own data?”

    Step 3: Prove it on the buyer’s own work One measure, the data and a decision date, agreed before the trial

  4. 4 “Can I believe these numbers?”

    Step 4: Check every claim before you publish A dated, repeatable test behind each one

  5. 5 “Which products do this?”, asked of an AI assistant

    Step 5: Show up in AI answers Plain pages on what it does, who it is for, the price and the limits

  6. 6 “Will the bill stay predictable?”

    Step 6: Price it to match the value A base fee with usage included, and a price that covers running costs

The six steps in this post. The buyer questions are my framing, not survey findings.

Step 1: Say the job first, then how the AI does it

Use this order: who it is for, the job it does, the result, then how AI does it, then the limit. The limit is not a weakness. It answers the first question a careful buyer will ask.

Example Co., step 1 Fictional

Who and job: owners of tutoring centers who need to show parents their child’s reading progress. Result: a weekly report for every family, without extra tutor time. How: the coach listens as a child reads aloud and flags words to practice. Limit: a tutor checks every flag before parents see the report. The new headline: “Show every parent their child’s reading progress every week, without extra tutor time.” All wording is fictional.

Figure 3 Example Co.’s new headline names the job and the result; the AI and its limit come underneath
  1. Who it is for
  2. The job
  3. The result
  4. How the AI does it
  5. The limit

Before Homepage headline

“AI-powered reading assessment platform”

  • Says how it works, not who it is for
  • Names no job and no result a buyer can check

After Homepage, top to bottom Fictional

  • Who Written for tutoring-center owners who report to parents
  • Job and result “Show every parent their child’s reading progress every week, without extra tutor time.”
  • How The coach listens as a child reads aloud and flags words to practice.
  • Limit A tutor checks every flag before parents see the report.

Example Co.’s homepage from Step 1, in the order the step sets out. All wording is fictional.

Do this

Rewrite your headline in that order, then test it with the Five-Second Message Test.

Step 2: Compete on what the model cannot give your rivals

Stop competing on the model. Stanford’s 2025 AI Index found that the price of using a model that matches GPT-3.5 on a standard benchmark fell more than 280-fold in under two years, from $20.00 per million tokens in November 2022 to $0.07 in October 2024[2]. What was an advantage can become a commodity within a year or two.

Compete on what is harder to copy: how well the product fits the buyer’s workflow, the domain data behind it, its measured accuracy on the buyer’s task, and the support that gets customers to results. And remember the new competitor: 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[4]. Your message has to answer “why not build it ourselves?”, and often “why not use a free chatbot?”

What Duolingo’s shareholder letter shows

When Duolingo launched its AI tier, it described the features by the job they do and placed its advantage around the model. Its first-quarter 2023 shareholder letter says the two AI features in Duolingo Max, “built with GPT-4”, “are aimed at teaching better by emulating a human tutor experience”. It adds: “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.’”[6] Its annual report for 2023 describes Max as a premium tier “priced higher than Super Duolingo”[7]. This is what the filings say, not an account of how Duolingo made its choices internally.

Figure 4 The model gets cheaper for everyone, so Duolingo and Example Co. put their advantage around it

Price of using a model that matches GPT-3.5 on the MMLU benchmark, per million tokens

November 2022 $20.00 October 2024Gemini-1.5-Flash-8B $0.07 280-fold cheaper

Duolingo First-quarter 2023 shareholder letter, quoted exactly

The advantage around it

“vast user data” “in-house machine learning and AI expertise”

The model “built with GPT-4”

“gamified approach” “brand”

“We’re not using GPT-4 ‘off the shelf.’”

Example Co. Fictional Its hard-to-copy list, from Step 2

The advantage around it

Accuracy tested on K–3 children reading aloud Built into the tutoring session

The model Models any rival can use too

A weekly progress report for parents Clear rules on children’s recordings

Competitor A and a free chatbot can use the same kind of model.

Show the numbers as a table
Date Price per million tokens Model
November 2022 $20.00 GPT-3.5 level
October 2024 $0.07 Gemini-1.5-Flash-8B

Stanford HAI, AI Index Report 2025 [2]: price of querying a model that scores the equivalent of GPT-3.5 (64.8) on MMLU; bars drawn to scale. Duolingo’s shareholder letter of May 9, 2023, filed with the SEC[6]: what the filing says, not how Duolingo decided. Example Co.’s list is fictional.

Example Co., step 2 Fictional

The team’s alternatives are a parent with a free chatbot, a tutor doing checks by hand, and Competitor A’s reading app. Its hard-to-copy list (fictional): accuracy tested on K–3 children reading aloud, a product built into the tutoring session, the weekly parent report, and clear rules on children’s recordings.

Do this

Write one sentence on why a buyer should choose you over building it, and one on why over a free chatbot.

Step 3: Prove it on the buyer’s own work

Let buyers see it on their own work. In Forrester’s 2026 research, drawn from a survey of nearly 18,000 business buyers, more than 60% reported buying some form of trial, yet just over a third said they planned to convert to a fully paid version with the same provider[8]. A trial is a test, not a sale.

Make trials convert by agreeing three things before they start: one workflow and one success measure (time, cost or error rate); the data the test will use, and how it is protected; and a decision date. Then let a person walk the buyer through the results. In a Gartner survey of 645 B2B buyers, 69% said they prefer to validate AI-generated insights with sales reps[9].

Example Co., step 3 Fictional

A 30-day pilot with one center, 20 readers and three tutors. The measure is tutor minutes spent on reading checks each week, before and after. Recordings are deleted at the end unless the center buys. On day 30 the owner decides: buy, extend once with a new date, or stop. All details are fictional.

Figure 5 Example Co. agrees the measure, the data and the decision date on day 0, and the owner decides on day 30
More than 60% of business buyers report buying some form of trial, from limited pilots to paid sandboxes[8] Just over a third said they planned to convert to a fully paid version with the same provider[8]

Example Co.’s 30-day pilot with one tutoring center Fictional

  1. Day 0

    Agree three things, in writing

    1. One measure Tutor minutes spent on reading checks each week, before and after
    2. The data 20 readers and three tutors; recordings deleted at the end unless the center buys
    3. A decision date Day 30

  2. Days 1 to 29

    The pilot runs

    Children read to the coach; a tutor checks every flag; tutor minutes are counted each week.

  3. Day 30

    Review the results with the owner, in person, then decide

    • Buy
    • Extend once, with a new date
    • Stop

Forrester, The State Of Business Buying, 2026 [8], from its survey of nearly 18,000 business buyers. Example Co.’s pilot, from Step 3, is fictional.

Do this

Put the decision date in writing before the pilot starts.

Step 4: Check every claim before you publish

Claim only what you can back with a test at the time you say it. In 2025 the FTC acted against Workado, which had marketed an AI-content detector as 98% accurate; the FTC’s complaint said independent testing found 53% accuracy on general-purpose content[3]. The final order requires “competent and reliable evidence” for future accuracy claims[10]. The cases continue: in May 2026, three firms agreed to pay $930,000 under proposed FTC orders over an “Active Listening” ad service sold as AI-powered; the FTC said it never listened to any conversations, and the firms were reselling email lists[11]. The SEC pursues what its then chair called “AI washing”[12]; in 2025 it found that Presto Automation had falsely claimed its drive-thru AI eliminated the need for human order-taking, when most orders still needed a person[13].

If you sell in the European Union, Article 50 of the AI Act has applied since August 2, 2026: people must be told when they are interacting with an AI system, unless that is obvious[14]. The Commission’s guidance, updated after the Act was amended in July 2026, keeps that date[15],[16]. If your product serves children under 13, US children’s privacy rules count a recording of a child’s voice as personal information[17]. This is general information, not legal advice.

Three quick checks for every claim:

  1. A number: is there a dated, repeatable test behind it?
  2. “Autonomous” or “agent”: does the product act without a person doing the work?
  3. “Secure” or “private”: can you name the standard, audit or policy?

Example Co., step 4 Fictional

An early deck said “95% accurate”, with no test behind it. After a proper test, the team writes: “In a March test on 500 recordings, the coach agreed with a trained tutor on 9 in 10 flagged words.” Because the product records children’s voices, it gets legal advice on children’s privacy before launch. The claim and its numbers are fictional.

Figure 6 Workado advertised 98% accuracy and testing found 53%; Example Co. puts every claim through three checks first

Workado’s AI Content Detector, telling AI-written from human-written text

Accuracy it advertised Its marketing claim

98%

Accuracy in independent testing On general-purpose content, per the FTC’s complaint

53%

50%: a coin toss. The FTC’s consumer protection director said the product “did no better than a coin toss”.

Three checks before any claim goes live

  1. A number

    Is there a dated, repeatable test behind it?

    If not Run the test, or drop the number

  2. “Autonomous” or “agent”

    Does it act without a person doing the work?

    If not Say what a person still does

  3. “Secure” or “private”

    Can you name the standard, audit or policy?

    If not Name it, or drop the word

Example Co.’s accuracy claim Fictional

Before

“95% accurate”

No test behind it: fails check 1

After

“In a March test on 500 recordings, the coach agreed with a trained tutor on 9 in 10 flagged words.”

Dated, repeatable and specific

Show the numbers as a table
Workado’s AI Content Detector Accuracy
Advertised 98%
Independent testing on general-purpose content, per the FTC’s complaint 53%

FTC press releases of April 28 and August 28, 2025[3],[10]; bars drawn to scale, 0 to 100%. The three checks are my practice. Example Co.’s claim, from Step 4, is fictional.

Do this

List every claim on your site with its evidence and date next to it. Delete what has none.

Free template

Check a launch against the rules before it goes live

The free Marketing Compliance Checklist covers nine rule sets to check before a campaign, email or AI feature goes live, each with a link to the regulator’s own page.

Get the checklist

Step 5: Show up in AI answers

Increasingly, buyers look in AI assistants. Forrester found 94% of business buyers use AI during their buying process[8]. If ChatGPT, Gemini or Perplexity cannot describe what you do, you can be missed early in the buyer’s research. Publish clear, factual pages on what the product does, who it is for, pricing and limits, so AI assistants can cite them. My post on why ChatGPT does not recommend your company covers this.

After that, I would favor hands-on channels: a free trial or sandbox, short demos on realistic data, and content that shows how you evaluate the product.

Example Co., step 5 Fictional

The team publishes one plain page with what the coach does, who it is for, the price and its limits, then asks three AI assistants to describe the product each month and fixes the page when an answer is wrong. It demos live at tutoring-center association meetings. Fictional choices.

Do this

Ask the AI assistants your buyers use to describe your product, and note what they get wrong.

Step 6: Price it to match the value

Match the price to the value and to your running costs. In Bain’s June 2026 analysis of public pricing at about 200 B2B software companies, about four in five vendors introducing AI pricing chose capacity models, where the customer commits to a set amount of AI use, such as a pool of credits. Of companies adding a metered AI charge, only about 10% charged for outcomes, such as a resolved customer issue[18]. A base fee plus included usage is a predictable starting point. The Marketing AI Products Playbook compares the options.

Example Co., step 6 Fictional

$200 a month per center, including up to 40 active readers, with extra readers at $4 each a month. Most centers get a predictable bill; the price grows only for large centers, where Example Co.’s running costs also grow. The reader limit and add-on price are fictional.

Figure 7 Example Co.’s bill stays at $200 up to 40 readers, then grows with use; few software vendors charge for outcomes

Example Co.’s monthly bill for one center, by active readers Fictional

$200 $200 $200 $280 $360 $440 40 readers included 10 20 40 60 80 100

Active readers

Base fee, $200, with up to 40 readers Extra readers, $4 each a month

How software vendors that add a metered AI charge set the meter, about 200 companies

35% 55% 10%
  1. Effort, about 35% Tokens, compute, agent hours
  2. Output, about 55% Drafts written, records updated, leads recommended
  3. Outcome, about 10% An issue resolved, revenue collected
Show the numbers as a table
Active readers Monthly bill (fictional)
$200
$200
$200
$280
$360
100 $440
AI meter (Bain, June 2026) Share
Effort: tokens, compute, agent hours About 35%
Output: drafts written, records updated, leads recommended About 55%
Outcome: an issue resolved, revenue collected About 10%

Example Co.’s prices, from Step 6, are fictional; columns drawn to scale. Bain & Company, analysis of public pricing at about 200 B2B software companies, June 2026[18]; shares are approximate.

Do this

Check what your price would be for your smallest and largest likely customer, and whether both cover your AI running costs.

What changes by stage?

  • Pre-seed — One buyer, one job, one measured result. Write the limits down.
  • Seed — A repeatable pilot with agreed measures, a security and data-use page, and pages AI assistants can cite.
  • Series A–B — Content for larger buying groups (security, legal, finance), a quarterly value report for customers, and a regular review of every public claim.

What mistakes should I avoid?

  • “AI-powered” as the headline. It says nothing a buyer can use.
  • Untested numbers. They create legal risk and break trust when buyers test them.
  • Free pilots with no end date. Without a decision date, a pilot drifts into free use.
  • Hiding the limits. Buyers find them anyway, usually in the trial.
  • Ignoring AI search. Buyers are asking assistants before they visit your site.

Related on AAJ

Articles

Playbooks

Templates and tools

References

The superscript numbers in the article match this list.

Regulators’ pages change; every FTC, SEC, European Commission and eCFR page above was checked on October 14, 2026.

  1. McClain, C. and Park, E. (2026). “Young adults in the U.S. are increasingly wary of AI, concerned it will take jobs.” Pew Research Center, short read, August 18, 2026. Survey of 3,488 US adults, June 22 to 28, 2026, with the same question asked each year since 2021. pewresearch.org
  2. Stanford Institute for Human-Centered Artificial Intelligence (2025). Artificial Intelligence Index Report 2025. The price of querying a model that scores the equivalent of GPT-3.5 (64.8) on MMLU fell from $20.00 per million tokens in November 2022 to $0.07 in October 2024 (Gemini-1.5-Flash-8B). hai.stanford.edu
  3. Federal Trade Commission (2025). “FTC Order Requires Workado to Back Up Artificial Intelligence Detection Claims.” Press release, April 28, 2025. ftc.gov
  4. McKinsey & Company and QuantumBlack, AI by McKinsey (2026). “The state of AI in 2026: On the road to ROI.” August 25, 2026. Online survey of 1,719 participants in 97 countries, May 4 to June 8, 2026. mckinsey.com
  5. 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. Experimental surveys with more than 1,000 US adults, as described in Washington State University’s press release of July 30, 2024. doi:10.1080/19368623.2024.2368040
  6. Duolingo, Inc. (2023). Shareholder letter for the first quarter of 2023, May 9, 2023, filed with the US Securities and Exchange Commission. sec.gov
  7. Duolingo, Inc. (2024). Annual report on Form 10-K for the fiscal year ended December 31, 2023, filed February 29, 2024. sec.gov
  8. Forrester (2026). “The State Of Business Buying: Risk-Averse Buyers Demand Proof, Not Promises.” Blog post on The State Of Business Buying, 2026, which draws on Forrester’s Buyers’ Journey Survey of nearly 18,000 business buyers worldwide. forrester.com
  9. Gartner (2026). “Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights.” Press release, May 20, 2026; survey of 645 B2B buyers, August to September 2025. gartner.com
  10. Federal Trade Commission (2025). “FTC Approves Final Order against Workado, LLC, Which Misrepresented the Accuracy of its Artificial Intelligence Content Detection Product.” Press release, August 28, 2025. ftc.gov
  11. Federal Trade Commission (2026). “FTC to Require Cox Media Group, Two Other Firms to Pay Nearly $1 Million to Settle Charges They Deceived Customers About ‘Active Listening’ AI-Powered Marketing Service.” Press release, May 21, 2026; proposed orders, open for public comment. ftc.gov
  12. US Securities and Exchange Commission (2024). “SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence.” Press release 2024-36, March 18, 2024. sec.gov
  13. US Securities and Exchange Commission (2025). Presto Automation Inc., settled administrative proceeding, January 14, 2025, File No. 3-22413. sec.gov
  14. Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 50(1), on AI systems that interact with people, and Article 113, which applies Article 50 from August 2, 2026. Official Journal of the European Union. eur-lex.europa.eu
  15. European Commission. “Transparency obligations under Article 50 of the AI Act” (FAQ), last updated July 24, 2026. Article 50 applies from August 2, 2026; a grace period to December 2, 2026 covers only the marking of AI-generated content by systems already on the market. digital-strategy.ec.europa.eu
  16. Regulation (EU) 2026/1744 of July 8, 2026, the Digital Omnibus on AI, amending Regulation (EU) 2024/1689. Official Journal of the European Union, July 24, 2026. eur-lex.europa.eu
  17. Code of Federal Regulations, 16 CFR 312.2, Children’s Online Privacy Protection Rule, definitions, as amended April 22, 2025: a child is under 13, and personal information includes an audio file that contains a child’s voice. ecfr.gov
  18. Maltiel, E. and Sandberg, J. (2026). “AI Pricing: A Reality Check on Effort, Usage, and Outcomes.” Bain & Company, August 10, 2026. Analysis of public pricing at about 200 B2B software companies, June 2026. bain.com

More in Industry Playbooks

Part of the Industry Playbooks hub - see all 10 resources on this topic.

Also useful in AI Search & Agent Readiness.