A buyer reads “99% accurate” on an AI landing page and asks one question: accurate at what, tested by whom? If the page cannot answer in the next sentence, the number hurts more than it helps.
AI claims draw scrutiny from buyers and regulators alike. This post shows how to write accuracy and trust claims on a landing page that you can stand behind: a four step check, weak claims rewritten, what regulators and design guides point to and a test for your own page.
Key takeaways
Trust on an AI page comes from claims a buyer can check, not from bigger numbers.
Use the Four S Check on every claim: Say it, Show it, Source it, Scope it.
Show the product’s limits on purpose. Google’s People + AI Guidebook argues users should calibrate trust, not trust blindly.
A claim you cannot support is a legal risk as well as a credibility risk. US regulators have kept pursuing AI claims since 2024.
One checkable claim beats five impressive ones.
What does trust mean on an AI landing page?
Trust on an AI landing page means a buyer believes your claims enough to take the next step, and has been given what they need to check them. It does not mean they believe the AI is perfect.
Google’s People + AI Guidebook frames trust in AI interfaces as calibration, a calibrated trust AI approach. Because AI products are built on statistics and probability, it says users should not trust the system completely, and should instead have the right level of trust given what it can and cannot do. That idea applies to marketing pages too. A page that promises perfection sets the buyer up to be disappointed, and a disappointed buyer remembers.
The practical consequence: a landing page that states what the AI does well, where it can be wrong and how the buyer stays in control is more persuasive to a careful buyer than a page that only celebrates.
What is the Four S Check?
The Four S Check is a short test for any accuracy or trust claim on an AI page. A claim passes when it can be Said plainly, Shown in the product, Sourced to evidence and Scoped to the conditions where it is true.
Step | The question | A passing claim | A failing claim |
|---|---|---|---|
Say | Can a buyer read it in one sentence? | “Flags unusual invoices for review” | “Next generation intelligent finance AI” |
Show | Can the buyer see it in the product? | A screen showing the flag and the reason | A stock image of a shield |
Source | Can the number be traced? | “Tested on 2,000 invoices from 14 customers, March 2026” | “98% accurate” |
Scope | Does it say where it holds? | “On typed English invoices under 5 pages” | No conditions stated |
The Four S Check targets the failure that repeats on AI pages: a confident number with no method behind it. Each step closes one gap. If a claim fails two or more steps, cut it or fix it before it ships.
Pro tip: Write the Scope line first. If you cannot say where a number holds, you do not yet know what it means, and neither will the buyer.
How do you show AI accuracy on a landing page?
Show AI accuracy on a landing page by writing each of your AI accuracy claims as a sentence with a number, a method, a date and a condition. Anything shorter is a slogan.
Weak claim | What is missing | Stronger version (template) |
|---|---|---|
“99% accurate” | Test, data, conditions | “[X]% agreement with human reviewers on [N] [documents] from [source], tested [month year]” |
“Enterprise grade AI” | Meaning | “SOC 2 Type II report available on request. Customer data not used to train models.” |
“Trusted by thousands” | Who, how many | “Used by [N] teams, including [named customer with permission]” |
“Hallucination free” | Almost certainly untrue | “Every answer links to the source passage so you can check it” |
“Replaces your team” | Overclaim | “Handles [named task] so your team can focus on [named work]” |
These are templates, not claims. Fill them only with numbers you can support. If you cannot, use the “Show” version instead, which describes a behavior of the product rather than a score.
What evidence do regulators expect for AI claims?
US regulators expect AI performance claims to be backed by evidence at the time they are made. The Federal Trade Commission launched Operation AI Comply on September 25, 2024, a sweep against companies it alleged made deceptive claims about AI, and law firm reviews in August 2026 say enforcement has continued two years on.
Two patterns from those reviews matter for landing pages. First, an “AI powered” label does not substantiate a claim. Second, a claim must hold up for the use cases the marketing implies, not only the one the model was built for.
This is not legal advice, and your counsel should review any performance claim. As a design and messaging matter, the lesson is simple: keep a short file behind every number on your page, with the test, the data and the date. If you cannot produce the file, take the number down.
Also read: AI Homepage Messaging Mistakes: 7 Fixes From Three Homepage Teardowns
Examples in the tables below are illustrations, not claims about any real product.
What does a good trust section look like?
The best AI landing page proof shows how the AI reaches an answer and lets the buyer check it. Abridge, a clinical documentation company, describes this approach in a post by its general counsel, dated November 12, 2025. It says draft notes are linked to the conversation they came from, so clinicians can check, challenge and decide whether to rely on them, and that the design is meant to be easy to audit. In the post, Abridge general counsel Tim Hwang writes that “Abridge-generated draft clinical documentation is traceable to the source conversation.”
What that does well: it turns a vague promise of trust into a visible mechanism. The reader can imagine clicking from a sentence to its source. What a landing page should add: a screen of the feature, so the claim is shown and not only described. The post describes the principle, and a visual would let a buyer see it in seconds.
Pro tip: Link a claim to its evidence on the page itself. A footnote that says “Methodology” and opens a short, dated explanation does more for trust than a larger badge.
Also read: AI Medical Scribe UX Teardown: What Abridge, Suki, Nuance DAX and Freed Get Right
What should an AI trust section include?
Include five elements, in this order, and keep each short.
One checkable headline claim. Say it, show it, source it, scope it.
A product view of the evidence. A screenshot of the source link, the citation or the audit trail.
A limits line. One sentence on where the AI is less reliable and what the user does then.
A control line. How the user reviews, edits or overrides the output.
A security and data line. Where data goes, whether it trains models, and which reports exist, stated as facts and with dates.
This order matters. A limits line placed right after a claim often raises credibility, because it signals the company knows its own product. Hiding limits can push careful buyers to look for them elsewhere, such as review sites and forums, where you do not control the wording.
What are the most common AI trust mistakes on landing pages?
A big number with no method. It invites suspicion, not belief.
A shield icon in place of evidence. Icons are decoration. Evidence is a link.
Claiming certainty. “Never wrong” is a promise one screenshot can break.
Hiding the limits. Careful buyers read the fine print and the review sites.
A trust page nobody can find. Put one proof line in the hero and link to the details.
Letting claims age. A 2024 test cited in 2026 is a stale claim. Date every figure and refresh it.
Also read: Product Messaging for AI Startups: What It Is and When to Hire Help
How do you test your own trust claims this week?
Run the Four S Check on every number and badge on your landing page, then fix or cut anything that fails two steps.
List every accuracy claim, number and badge on the page.
Score each on Say, Show, Source and Scope, one point each.
Fix anything under three points. Remove anything under two.
Add a dated methodology link to the headline claim.
Show the page to three buyers and ask which claim they believe most and why.
If the claim they believe most is the one with a method behind it, the page is doing its job.
To see how we work, read our Nura healthcare platform dashboard and ORYN AI agent monitoring landing page case studies, or browse all of our case studies.
Keep reading: How to Design Trust Into AI Interfaces: 8 Moments Where Users Decide, Digital Health SaaS Messaging: How to Write for Clinicians, Admins and Patients, SaaS Website Messaging Audit: A 20 Point Checklist You Can Run in an Afternoon and AI Landing Page Design: Examples and Lessons From Three Pages We Designed
Can every claim on your AI page pass the Four S Check?
Send us your landing page and we will score each claim and show you which to fix first. Get a trust claims review · See our work
AI accuracy and trust FAQs
State one claim with a number, the test behind it, the date and the conditions where it holds, and link to a short methodology. Then show the feature in the product, such as a source link or an audit trail. A buyer who can see how the number was produced is far more likely to believe it.
Publish a number only if you can support it with a test you can describe. A modest, well explained figure is more persuasive than a high one with no method. If you cannot support a number yet, describe a behavior instead, such as “every answer links to its source,” and add the number when you have the evidence.
Calibrated trust means a user trusts the system as much as it deserves, no more and no less. Google’s People + AI Guidebook recommends helping users calibrate their trust by explaining what the system can and cannot do. For landing pages, that means showing strengths and limits together.
Yes. Kraftbase is a product messaging and UI/UX design studio for AI and B2B SaaS companies, working as an AI product design agency, and trust sections are part of the homepage and product page work. We start with the claims you can support and design the evidence into the page.
Yes. The US Federal Trade Commission has pursued companies over unsubstantiated AI claims since it launched Operation AI Comply in 2024, and reviews published in 2026 describe continued enforcement. This is general information, not legal advice. Have counsel review performance claims, and keep evidence on file for every number you publish.
Yes, briefly. One line on where the AI is less reliable and what the user does then tends to build credibility with careful buyers. Place it near the main claim, and pair it with a line about user control, such as reviewing or editing outputs.
Linking each AI output to its source. Abridge describes draft clinical notes that trace back to the conversation they came from, so clinicians can check and decide whether to rely on them. The pattern works in any domain: let the user click from the answer to the evidence.
About the author
Jenish is a designer at Kraftbase, a product messaging and UI/UX design studio in Vadodara, India, that helps AI and B2B SaaS teams in the US, UAE and India explain and design what they’ve built.
Sources
People + AI Guidebook, “Explainability + Trust,” Google PAIR
FTC, “FTC Announces Crackdown on Deceptive AI Claims and Schemes,” Sept 25, 2024
Operation AI Comply launch coverage, Davis Polk client update, Oct 2024
FTC, case page for DoNotPay, part of Operation AI Comply, ftc.gov
“Operation AI Comply 2 Years Later,” Holland & Knight, Aug 18, 2026
Tim Hwang, “Building trusted healthcare AI,” Abridge, Nov 12, 2025
Trust claims review
Can every claim on your AI page pass the Four S Check?
Send us your landing page and we will score each claim and show you which to fix first.
A buyer reads “99% accurate” on an AI landing page and asks one question: accurate at what, tested by whom? If the page cannot answer in the next sentence, the number hurts more than it helps.
AI claims draw scrutiny from buyers and regulators alike. This post shows how to write accuracy and trust claims on a landing page that you can stand behind: a four step check, weak claims rewritten, what regulators and design guides point to and a test for your own page.
Key takeaways
Trust on an AI page comes from claims a buyer can check, not from bigger numbers.
Use the Four S Check on every claim: Say it, Show it, Source it, Scope it.
Show the product’s limits on purpose. Google’s People + AI Guidebook argues users should calibrate trust, not trust blindly.
A claim you cannot support is a legal risk as well as a credibility risk. US regulators have kept pursuing AI claims since 2024.
One checkable claim beats five impressive ones.
What does trust mean on an AI landing page?
Trust on an AI landing page means a buyer believes your claims enough to take the next step, and has been given what they need to check them. It does not mean they believe the AI is perfect.
Google’s People + AI Guidebook frames trust in AI interfaces as calibration, a calibrated trust AI approach. Because AI products are built on statistics and probability, it says users should not trust the system completely, and should instead have the right level of trust given what it can and cannot do. That idea applies to marketing pages too. A page that promises perfection sets the buyer up to be disappointed, and a disappointed buyer remembers.
The practical consequence: a landing page that states what the AI does well, where it can be wrong and how the buyer stays in control is more persuasive to a careful buyer than a page that only celebrates.
What is the Four S Check?
The Four S Check is a short test for any accuracy or trust claim on an AI page. A claim passes when it can be Said plainly, Shown in the product, Sourced to evidence and Scoped to the conditions where it is true.
Step | The question | A passing claim | A failing claim |
|---|---|---|---|
Say | Can a buyer read it in one sentence? | “Flags unusual invoices for review” | “Next generation intelligent finance AI” |
Show | Can the buyer see it in the product? | A screen showing the flag and the reason | A stock image of a shield |
Source | Can the number be traced? | “Tested on 2,000 invoices from 14 customers, March 2026” | “98% accurate” |
Scope | Does it say where it holds? | “On typed English invoices under 5 pages” | No conditions stated |
The Four S Check targets the failure that repeats on AI pages: a confident number with no method behind it. Each step closes one gap. If a claim fails two or more steps, cut it or fix it before it ships.
Pro tip: Write the Scope line first. If you cannot say where a number holds, you do not yet know what it means, and neither will the buyer.
How do you show AI accuracy on a landing page?
Show AI accuracy on a landing page by writing each of your AI accuracy claims as a sentence with a number, a method, a date and a condition. Anything shorter is a slogan.
Weak claim | What is missing | Stronger version (template) |
|---|---|---|
“99% accurate” | Test, data, conditions | “[X]% agreement with human reviewers on [N] [documents] from [source], tested [month year]” |
“Enterprise grade AI” | Meaning | “SOC 2 Type II report available on request. Customer data not used to train models.” |
“Trusted by thousands” | Who, how many | “Used by [N] teams, including [named customer with permission]” |
“Hallucination free” | Almost certainly untrue | “Every answer links to the source passage so you can check it” |
“Replaces your team” | Overclaim | “Handles [named task] so your team can focus on [named work]” |
These are templates, not claims. Fill them only with numbers you can support. If you cannot, use the “Show” version instead, which describes a behavior of the product rather than a score.
What evidence do regulators expect for AI claims?
US regulators expect AI performance claims to be backed by evidence at the time they are made. The Federal Trade Commission launched Operation AI Comply on September 25, 2024, a sweep against companies it alleged made deceptive claims about AI, and law firm reviews in August 2026 say enforcement has continued two years on.
Two patterns from those reviews matter for landing pages. First, an “AI powered” label does not substantiate a claim. Second, a claim must hold up for the use cases the marketing implies, not only the one the model was built for.
This is not legal advice, and your counsel should review any performance claim. As a design and messaging matter, the lesson is simple: keep a short file behind every number on your page, with the test, the data and the date. If you cannot produce the file, take the number down.
Also read: AI Homepage Messaging Mistakes: 7 Fixes From Three Homepage Teardowns
Examples in the tables below are illustrations, not claims about any real product.
What does a good trust section look like?
The best AI landing page proof shows how the AI reaches an answer and lets the buyer check it. Abridge, a clinical documentation company, describes this approach in a post by its general counsel, dated November 12, 2025. It says draft notes are linked to the conversation they came from, so clinicians can check, challenge and decide whether to rely on them, and that the design is meant to be easy to audit. In the post, Abridge general counsel Tim Hwang writes that “Abridge-generated draft clinical documentation is traceable to the source conversation.”
What that does well: it turns a vague promise of trust into a visible mechanism. The reader can imagine clicking from a sentence to its source. What a landing page should add: a screen of the feature, so the claim is shown and not only described. The post describes the principle, and a visual would let a buyer see it in seconds.
Pro tip: Link a claim to its evidence on the page itself. A footnote that says “Methodology” and opens a short, dated explanation does more for trust than a larger badge.
Also read: AI Medical Scribe UX Teardown: What Abridge, Suki, Nuance DAX and Freed Get Right
What should an AI trust section include?
Include five elements, in this order, and keep each short.
One checkable headline claim. Say it, show it, source it, scope it.
A product view of the evidence. A screenshot of the source link, the citation or the audit trail.
A limits line. One sentence on where the AI is less reliable and what the user does then.
A control line. How the user reviews, edits or overrides the output.
A security and data line. Where data goes, whether it trains models, and which reports exist, stated as facts and with dates.
This order matters. A limits line placed right after a claim often raises credibility, because it signals the company knows its own product. Hiding limits can push careful buyers to look for them elsewhere, such as review sites and forums, where you do not control the wording.
What are the most common AI trust mistakes on landing pages?
A big number with no method. It invites suspicion, not belief.
A shield icon in place of evidence. Icons are decoration. Evidence is a link.
Claiming certainty. “Never wrong” is a promise one screenshot can break.
Hiding the limits. Careful buyers read the fine print and the review sites.
A trust page nobody can find. Put one proof line in the hero and link to the details.
Letting claims age. A 2024 test cited in 2026 is a stale claim. Date every figure and refresh it.
Also read: Product Messaging for AI Startups: What It Is and When to Hire Help
How do you test your own trust claims this week?
Run the Four S Check on every number and badge on your landing page, then fix or cut anything that fails two steps.
List every accuracy claim, number and badge on the page.
Score each on Say, Show, Source and Scope, one point each.
Fix anything under three points. Remove anything under two.
Add a dated methodology link to the headline claim.
Show the page to three buyers and ask which claim they believe most and why.
If the claim they believe most is the one with a method behind it, the page is doing its job.
To see how we work, read our Nura healthcare platform dashboard and ORYN AI agent monitoring landing page case studies, or browse all of our case studies.
Keep reading: How to Design Trust Into AI Interfaces: 8 Moments Where Users Decide, Digital Health SaaS Messaging: How to Write for Clinicians, Admins and Patients, SaaS Website Messaging Audit: A 20 Point Checklist You Can Run in an Afternoon and AI Landing Page Design: Examples and Lessons From Three Pages We Designed
Can every claim on your AI page pass the Four S Check?
Send us your landing page and we will score each claim and show you which to fix first. Get a trust claims review · See our work
AI accuracy and trust FAQs
State one claim with a number, the test behind it, the date and the conditions where it holds, and link to a short methodology. Then show the feature in the product, such as a source link or an audit trail. A buyer who can see how the number was produced is far more likely to believe it.
Publish a number only if you can support it with a test you can describe. A modest, well explained figure is more persuasive than a high one with no method. If you cannot support a number yet, describe a behavior instead, such as “every answer links to its source,” and add the number when you have the evidence.
Calibrated trust means a user trusts the system as much as it deserves, no more and no less. Google’s People + AI Guidebook recommends helping users calibrate their trust by explaining what the system can and cannot do. For landing pages, that means showing strengths and limits together.
Yes. Kraftbase is a product messaging and UI/UX design studio for AI and B2B SaaS companies, working as an AI product design agency, and trust sections are part of the homepage and product page work. We start with the claims you can support and design the evidence into the page.
Yes. The US Federal Trade Commission has pursued companies over unsubstantiated AI claims since it launched Operation AI Comply in 2024, and reviews published in 2026 describe continued enforcement. This is general information, not legal advice. Have counsel review performance claims, and keep evidence on file for every number you publish.
Yes, briefly. One line on where the AI is less reliable and what the user does then tends to build credibility with careful buyers. Place it near the main claim, and pair it with a line about user control, such as reviewing or editing outputs.
Linking each AI output to its source. Abridge describes draft clinical notes that trace back to the conversation they came from, so clinicians can check and decide whether to rely on them. The pattern works in any domain: let the user click from the answer to the evidence.
About the author
Jenish is a designer at Kraftbase, a product messaging and UI/UX design studio in Vadodara, India, that helps AI and B2B SaaS teams in the US, UAE and India explain and design what they’ve built.
Sources
People + AI Guidebook, “Explainability + Trust,” Google PAIR
FTC, “FTC Announces Crackdown on Deceptive AI Claims and Schemes,” Sept 25, 2024
Operation AI Comply launch coverage, Davis Polk client update, Oct 2024
FTC, case page for DoNotPay, part of Operation AI Comply, ftc.gov
“Operation AI Comply 2 Years Later,” Holland & Knight, Aug 18, 2026
Tim Hwang, “Building trusted healthcare AI,” Abridge, Nov 12, 2025
Trust claims review
Can every claim on your AI page pass the Four S Check?
Send us your landing page and we will score each claim and show you which to fix first.