Buyers increasingly open ChatGPT, Gemini or Perplexity and type something like "what are the best tools for X" or "alternatives to [competitor] for a 50-person team." The answer is a shortlist. If your product is not on it, you never enter the evaluation, no matter how good your demo is.
That changes what SEO work looks like for SaaS. Ranking a blog post is no longer the whole job. You need pages that answer buyer questions directly, in language a model can lift, and you need to let AI search crawlers reach them.
Here is a practical order of operations.
First, check whether AI search can reach you at all
This is the cheapest step and the one most teams skip.
In our study of 210 well-known sites (published 2026-09-21), we requested each site's robots.txt with a declared bot user agent and evaluated the rules with an RFC 9309 parser. Of the 166 readable files, 19.9% blocked at least one AI search or live-fetch crawler. Among SaaS sites specifically (n=48), 13% blocked at least one AI training crawler and 0% blocked any AI search crawler.
That 0% is the useful part for SaaS. In our sample, SaaS companies are not shutting AI search out. If your site does block one, you are the outlier, and it is probably accidental.
Accidental blocks look like this: a security plugin or CDN template ships with a robots.txt that disallows a long list of bots, nobody reviews it, and OAI-SearchBot or Claude-SearchBot ends up in the list. It is worth ten minutes to confirm.
What the vendor documentation says about the crawlers that matter for search visibility:
- OpenAI: OAI-SearchBot surfaces websites in ChatGPT search. Sites opted out of OAI-SearchBot are not shown in ChatGPT search answers. robots.txt changes can take about 24 hours. GPTBot is separate and collects content for model training; blocking it does not affect ChatGPT search. ChatGPT-User handles user-initiated actions and is not used to decide whether content appears in search.
- Anthropic: Claude-SearchBot indexes content to improve search results, and blocking it prevents indexing for search. Claude-User fetches pages when a user asks Claude a question; blocking it prevents retrieval for user queries. ClaudeBot is the training crawler. Anthropic says its bots honor robots.txt, support Crawl-delay, and it publishes its IP ranges.
- Perplexity: PerplexityBot surfaces and links sites in Perplexity search results and is not used to crawl content for foundation models. Perplexity-User fetches pages for user questions and generally ignores robots.txt because the user requested it. Changes may take up to 24 hours.
- Google: Google-Extended controls training and grounding for Gemini, not inclusion in Google Search, and it is not a ranking signal. AI Overviews are part of Google Search and governed by normal Googlebot crawling and indexing.
Two practical notes. robots.txt expresses intent, not access control, and sites can also block at the CDN or edge. In our sample, 44 of 210 sites refused or timed out for our declared bot, which is a different kind of gatekeeping entirely.
If you want to know what your current setup produces, our free AI visibility check generates three buyer questions for your brand and category, asks Perplexity with web search, and shows whether your brand is named, which competitors are named, and which sources get cited. Three checks per day. It does not track over time, so treat it as a snapshot, not a dashboard.
Build the pages AI answers pull from
Once crawlers can reach you, the work is editorial. Models assemble shortlists from pages that clearly state what a product is, who it is for, and how it compares. Five page types do most of that work for SaaS.
Category pages that name the category
A homepage that says "the platform for modern teams" gives a model nothing to file. A category page that says "project management software for agencies" gives it a slot to put you in.
Write one page per category you genuinely compete in, and say the category phrase in the H1, the first paragraph, and the page title. If you sell into two categories, build two pages. Do not make a model guess which one you belong to.
Integration pages
Buyers ask things like "what works with HubSpot and Snowflake." Integration pages answer that in a form a model can quote: what the integration does, what it syncs, which direction, and any limits. One page per meaningful integration beats a single grid of logos, because a grid has no sentences to extract.
Pricing pages with actual numbers
"Contact us for pricing" is invisible to a shortlist question that includes a budget. You do not have to publish enterprise tiers. You do have to publish something concrete: a starting price, a per-seat range, a free tier, or a clearly stated usage metric. If pricing is genuinely custom, say what drives the price and give a typical range where you can.
Comparison and alternatives pages
These are the highest-intent pages in B2B SaaS, and they are also the ones AI answers lean on when someone asks for alternatives to a named competitor. Write them honestly. A comparison page that claims you win every row reads as marketing and gets discounted. A page that states where a competitor is stronger, and for whom, is more likely to be treated as a source.
A short, factual "what is [product]" page
One page, plain sentences, no metaphor. What it does, who uses it, what it replaces, what it does not do. This is the page a model quotes when it introduces you in an answer.
Make third-party presence match your own claims
AI answers cite sources, and those sources are often not your site. Review sites, community threads, analyst write-ups, and press coverage all feed the picture. If your site says one thing and every third-party profile says another, the model has to choose, and it will usually trust the consensus over the vendor.
For B2B SaaS, the practical moves are unglamorous:
- Keep your profiles on major review platforms current: category, company size served, pricing model, integrations. Stale categories are worse than missing ones.
- Ask customers to describe outcomes in their own words. A review that says "cut onboarding from three weeks to four days" is quotable; "great product, easy to use" is not.
- Get into the roundups and listicles that already rank for your category. Those pages are frequently the cited source in AI answers, so being present on them matters more than being present on your own blog.
- Answer real questions in the communities your buyers use. Public, indexable answers are part of the corpus.
One caution: you cannot guarantee a mention, and no tool can. What you control is whether the underlying facts are available, consistent, and easy to quote.
Check the plumbing, then the content
A short audit you can run this week:
- Fetch your robots.txt and confirm OAI-SearchBot, Claude-SearchBot, and PerplexityBot are not disallowed. Check the CDN or edge rules too, since robots.txt is only part of the picture.
- List the buyer questions you would want to be shortlisted for. Ten is enough.
- For each one, ask whether you have a page that answers it in plain sentences. Mark the gaps.
- Fix the two highest-intent gaps first. Usually that is a comparison page and a pricing page with a real number on it.
- Run the AI visibility check to see which competitor names and sources come back for your category, and compare that source list against where you actually appear.
Start with one question
Pick the single buyer question that matters most to your pipeline, write down the five products you would expect an AI to name, and run that question through the visibility check. If you are not in the answer, look at which pages were cited. That gap, not a general content calendar, tells you what to build next.
One more thing worth knowing: robots.txt follows the most specific group that names a crawler, and within a group the longest matching rule wins. A stray Disallow: / in a named group for a search bot quietly removes you from the answers, and nobody gets an error message.