Summarize with AI
A landing page FAQ section for an AI product should answer the three questions buyers ask before they trust a model with their work: what happens when it gets something wrong, whether you train on their data, and what happens if OpenAI ships the same thing. Give each one two plain sentences, put them first, and move the billing questions to the pricing page.
This matters more for an AI startup than for any other kind of software. Those three questions decide the deal, and today they get answered only on your intro calls, by you. The FAQ is also the text an AI assistant is most likely to quote when someone asks it about you, so whatever you write there is the answer a stranger gets when you are not in the room.
Six closed rows about billing, and nothing about the model
Picture an AI code reviewer for engineering teams of ten to eighty people. This is an example, not a client. It reads every pull request, leaves comments on the lines that look risky, and learns which comments a team tends to ignore. The landing page is good.
Near the bottom, just above the footer, sits a section headed Frequently asked questions. Six accordion rows, all closed. Is there a free trial? Can I cancel anytime? Which languages do you support? Do you offer startup discounts? Is my code secure? How do I contact support? Open the fifth row and it says: Security is our top priority. We use industry-standard encryption.
Nothing on the page is false. But a head of engineering deciding whether to let this tool comment on her team's code is not wondering about cancellation. She is wondering what happens when the reviewer waves through a change that takes down checkout, whether her company's code ends up in someone's training set, and whether this whole product becomes a checkbox inside GitHub next year. None of those words appear anywhere on the page.
The three questions from every intro call
The founder knows those questions by heart, because he hears them on almost every call. He keeps a doc called Intro calls with a section for each prospect. Scroll through it and the same three lines keep showing up in his own shorthand: what if it's wrong, train on our code?, what if OpenAI does this.
He has a saved text snippet for the second one and pastes it into follow-up emails. He answers the third one differently on every call. When a lead goes quiet, he reads it as a pricing problem and tries a longer trial. The answers that would have moved the deal live in his head, in a snippet, and in a call recording nobody else will play. That is the whole wound in miniature. He knows his product, and the page does not say what he knows.
Where the trial goes quiet
Here is how the cost shows up. The head of engineering likes the trial and forwards the landing page to her CTO with one line: worth a look? The CTO opens it on his phone, scrolls to the bottom, taps Is my code secure?, and reads about industry-standard encryption. That does not answer whether the code is kept or used for training. He replies to her with the question. She emails the founder. Two or three days pass in email, the trial clock keeps running, and the team's attention moves on.
Measured on SaaS and product-led companies, opt-in free trials convert at around 18 percent. Most of the rest never say why they left, so a question left unanswered during a trial is an easy way to lose one, and a hard one to see. The third question costs more than a deal. Seed-stage AI startups raise at valuations roughly 42 percent above non-AI peers, and that gap assumes there is something behind the model. A founder who cannot answer what if OpenAI ships this in two sentences gets priced as the thin layer he is trying not to be.
Billing questions belong on the pricing page
Most FAQ advice was written for ordinary software. It says to list the questions that stop a signup: price, trial, cancellation, support. For an AI product the main risk is different. The buyer is worried about giving a model access to her work, and a trial question does nothing for that worry.
So here is the claim we would defend: the landing page FAQ should not hold a single billing question. Is there a free trial and Can I cancel belong on the pricing page, next to the plan they describe, where the reader who cares about them is already looking. The landing page FAQ is the one place on the page where you can name an objection in the buyer's own words without sounding defensive, because the format itself says people ask this. Spend it on the three questions that decide the deal.
There is a second reason. When someone asks an assistant whether the tool trains on customer code, the assistant answers from whatever it can find in writing. A question used as a heading with a two sentence answer under it is about the easiest text on the web to quote correctly. We wrote about how that changes your numbers in why a top position now gets you zero clicks. If the only public answer is industry-standard encryption, that vague line is what gets repeated.
Two sentences per objection, with one thing a reader can check
Each answer follows the same shape. The first sentence answers the question directly, yes or no, using the buyer's noun. The second sentence gives one specific thing that proves it: a setting, a contract term, a screen. Here are the three for the example code reviewer.
What happens when it approves something that breaks production? It cannot approve or merge anything, it only comments, and a person on your team still clicks Approve. Every comment shows the lines it read and a confidence label, and you can switch any rule off for a repository in one click.
Do you train on our code? No, your code is read to write the review and is not used to train any model, ours or our provider's. The data page lists every company that touches your code and how long each one keeps it.
What if OpenAI ships this? A general model can read a diff, but it cannot see which comments your team has ignored for the last year or which rules you switched off. The reviewer uses that history as context on every pull request, and it stays in your account, so the product gets more useful the longer you use it.
Notice that the third answer is careful not to contradict the second. This is the most common mistake we see when founders write these. The moat answer says the product learns from you, the data answer says your code is never used, and a careful reader spots the conflict. Write all three together and read them as a set.
The strongest second sentence is a measured result you show on the page itself, not one you describe. On the Chariot website, a site we designed for a speech reasoning lab, the model's quality is shown as blind human ratings against other systems and a word error rate, rather than a sentence calling it accurate. If you have numbers like that, point the FAQ answer at them. If the full data answer needs more than two sentences, it belongs on a trust page; our guide to the trust page buyers check covers what goes there. The FAQ links to it, and does not try to replace it.
Where the section sits, and what the markup still does
Put the FAQ section directly under the product demo or the proof section, before pricing, not after the final call to action. The visitor who has just watched the product work is the one with these questions. By the footer, she has already decided.
Order the rows by how often you hear them, not by how comfortable they are to answer. Open the first row by default so a scrolling reader sees one full answer without tapping. Make sure every answer is in the page's HTML when it loads, not fetched when someone opens the row, so crawlers and assistants can read it.
Then be realistic about FAQPage structured data. Since August 2023, Google has said FAQ rich results will only be shown for well-known, authoritative government and health websites. Adding the markup does no harm, and it is a tidy way to label questions and answers, but do not expect expandable questions under your search result. The visible text on the page is what does the work.
An afternoon to rewrite yours this week
- Open your last ten intro call notes or recordings and search for wrong, mistake, train, data, OpenAI, GitHub and the name of any big platform in your space. Write down each question in the buyer's words.
- Do the same with your sent folder. Any paragraph you have pasted into more than two follow-up emails is an FAQ answer that never made it to the page.
- Pick the three questions that come up most. For most AI products they will be the error question, the data question and the platform question.
- Write two sentences for each. Answer first, then one specific thing a reader can check. Read all three together and fix anything that contradicts.
- Move every billing question to the pricing page, next to the plan it describes.
- Place the section under the demo, open the first row, and check that the answers show up when you view the page source.
- Send the section to someone outside the company and ask them, from memory, what happens when the product is wrong. If they cannot say, the first sentence is not plain enough.
If you are hiring someone to rebuild the whole page around these answers, our ranking of landing page design agencies for AI startups is a place to start the shortlist.
Studio Maydit designs websites for AI founders in the US, UK and Europe, and the work usually starts where this post does: with what you say on calls that the page never says. We read your call notes, find the objections that stall deals, and build a page where a stranger gets your answers without you there, in Framer, Webflow or custom code, as a fixed scope project of three to four weeks. If your FAQ is still about free trials, book a 30 minute call and bring your last three call notes.
