Make pages that are easy to retrieve and easy to quote, and be present on the third-party sources the answers lean on. No single tactic guarantees a citation.
For SaaS teams whose buyers now ask AI tools for recommendations. You will leave with a readiness check and a short list of fixes.
A buyer asks an AI assistant which tool to use for a problem you solve. The answer names three products and explains why. Either you are in that answer or you are not, and the buyer usually never visits a page to find out what they missed. This guide covers what you can control about whether you are in it.
1Understand what an answer is made from
Most AI search tools do two things. They retrieve passages from a web index or a live fetch, then a language model writes an answer from those passages and cites some of them. So your page has to pass four tests in order: it can be fetched, it can be understood, a passage from it can stand alone, and the source looks trustworthy enough to cite.
- Fetched
- Understood
- Quoted alone
- Trusted
Most teams fail the first test and never know it. A page that renders its content with JavaScript, sits behind an over-eager bot rule, or is not indexed cannot be used however well it is written. Start there, because it is the cheapest fix and often the largest.
2Make sure the bots can read you
Different bots do different jobs, and you can treat them differently. Training crawlers collect content to train models. Search crawlers build an index for AI search. User-triggered fetchers read a page when a person asks about it. Blocking the first is a legitimate choice. Blocking the others removes you from answers you may want to appear in.
| Kind | Example user agents | What blocking does |
|---|---|---|
| Training | GPTBot, Google-Extended, ClaudeBot | Keeps content out of future training. Does not remove you from search answers. |
| Search index | OAI-SearchBot, PerplexityBot | Can keep you out of that product's answers. |
| User-triggered | ChatGPT-User | Stops the assistant reading your page when someone asks about you. |
User agent names change, so check each vendor's own documentation before relying on this list. A robots.txt that allows search bots and disallows a training bot looks like this:
# Allow AI search and user-requested fetches
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
# Opt out of model training
User-agent: GPTBot
Disallow: /
Sitemap: https://example.com/sitemap.xmlThen test what a simple crawler sees. Fetch the page without running scripts and look for your main heading and first paragraph:
curl -s https://example.com/pricing | grep -i "<h1"If the heading is missing, the content is injected by JavaScript. Ask your developers for server rendering or static generation on content pages so the text is in the first response.
3Write sections that stand alone
A retrieval system pulls out passages, not whole pages. A passage that depends on the one before it, or opens with three sentences of scene-setting, is hard to use. Make each section answer one question, put the answer first, then add the detail.
Hard to quote
Pricing is something we have thought a lot about. Every team is different, so we built a flexible approach that grows with you and reflects the value you get.
Easy to quote
Fernhollow charges per active seat, billed monthly or annually. The Team plan covers up to 25 seats and includes unlimited reports. Extra seats are added at the same per-seat price.
The second version names the product, states the model, the limit and what is included, and makes sense if copied out of the page.
- Write the heading as the question a buyer would type
- Answer it in the first one or two sentences, using the product's name, not a pronoun
- State when the answer changes, such as by plan or team size
- Support it with one sourced fact or a short example
4State the facts buyers ask about
When an assistant is asked what a tool costs, whether it integrates with another, or how secure it is, it looks for a page that says so plainly. If the facts are only in a PDF, an image, a pricing widget or a sales call, there is nothing to cite, and the answer may come from an old listing instead.
Facts to state in plain text, with a date
- Plan names, prices and billing terms
- Limits per plan: seats, usage, projects
- Integrations, one page each
- Security and compliance practices you actually have
- Setup time and who does it
- Who the product is for, and who it is not for
5Show who stands behind the page
Systems and readers both weigh who is speaking. A page with a named author who has relevant experience, a company page that says who you are, and facts that agree across your site and profiles reads as more reliable than anonymous copy.
- Name an author on guides and comparisons, with a short, real bio and a link
- Keep an About page with the company's name, location and what it does
- Use the same company description on your site, LinkedIn, review profiles and directories
- Mark up the organisation with its official profiles so the connection is explicit
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Fernhollow",
"url": "https://example.com",
"description": "Analytics software for independent retailers.",
"sameAs": [
"https://www.linkedin.com/company/fernhollow",
"https://github.com/fernhollow"
]
}6Be present where the answers look
Answers often cite sources other than the vendor: review sites, directories, respected publications and community discussions. If those sources describe your category and leave you out, or describe you wrongly, your own pages are fighting uphill.
- Run your buyer questions and write down the domains cited, ignoring your own
- For each, check whether you appear and whether the details are right
- Claim and complete profiles; correct wrong pricing or features
- Ask satisfied customers for honest reviews. Never buy or invent them
- Share real expertise in communities and publications your buyers read
7Measure it, honestly
Answers vary from run to run, so one check proves nothing. Keep a fixed set of buyer questions, run each several times in the tools your buyers use, and record whether you are named, cited and linked, and who is cited instead.
Add AI referral visits in your analytics and a how-did-you-hear question at signup. Treat the tracked referrals as a minimum, because many AI-influenced visits arrive as direct traffic.
Worked example
Fernhollow's first month (illustrative)
Fernhollow is a made-up analytics tool for independent retailers. The team writes twelve buyer questions, runs each three times in two assistants, and finds it is never cited. Raw-HTML checks show the pricing page loads its plans with JavaScript, and robots.txt blocks every AI bot because of an old firewall rule.
They fix access first, move the plan details into the HTML, rewrite five sections to lead with the answer, correct a wrong price on a review site and publish a comparison page. A month later they rerun the same twelve questions. They are cited in some runs for the comparison and pricing questions, and still absent on the broad category questions, which tells them where to work next.
8Your first month
- Week 1: write the question set, check access and fix anything that blocks bots
- Week 2: state your key facts in plain text with dates, and rewrite your five most important sections
- Week 3: add authors, align your company description and fix the profiles that cited sources rely on
- Week 4: rerun the question set, compare with week 1, and choose the next three fixes
Say it in 30 seconds
The 30-second version
Make sure bots can read your pages, then write each section so it opens with a direct answer. State facts such as pricing and limits plainly and date them, show who wrote the page, and be accurately present on the sources AI answers already cite. Measure with a fixed set of questions.
Worksheet
AI citation readiness worksheet
The full guide and a worksheet to fill in, as a PDF you can print or share with your team. The worksheet covers:
- Write down the ten questions a buyer asks before choosing a tool like yours
- For each question, note the page that should answer it, or mark it missing
- Fetch each page's raw HTML and confirm the answer is in it without JavaScript
- Check robots.txt for rules on GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Google-Extended and ClaudeBot
- and 8 more
One email with the PDF. No list, no follow-ups.
