How to get your SaaS recommended by ChatGPT and AI search (2026)

Try something before you read the rest of this. Open ChatGPT, or Claude, or Perplexity, and type the question your customer would type on a bad Tuesday. Not your brand name. The problem. "What's a good tool for chasing unpaid invoices?" or "cheapest way to add a help centre to a small SaaS."

If your product came up, good. If three competitors came up and you did not, you are in the same spot as most small teams we talk to, and the fix is less mysterious than the agencies selling "AI SEO" packages would like you to think.

This guide is for founders and tiny marketing teams: how assistants pick products, the technical checks that quietly keep sites out, the pages that get quoted, and how to tell whether it is working.

How an assistant decides which tools to name

An AI assistant can know about your product in two ways, and they behave very differently.

The first is memory. If your product was discussed enough before the model's training cutoff, it may simply know you exist. You cannot change that quickly. It shifts when new models ship, and it rewards products that were already talked about.

The second is retrieval, and this is where a small team has a real chance. For questions that need current information, the assistant runs a search, reads a handful of pages, and writes its answer from what it found. ChatGPT search, Perplexity, Claude with web search and Google's AI features all work some version of this way.

So the useful question is not "how do I get into the AI." It is two smaller, more practical questions:

Can the assistant's crawler reach and read your pages at all?
When it searches your category, does it land on pages that mention you?

Everything below is an answer to one of those two.

Step 1: Make sure the crawlers can actually get in

This is the boring part, and it is where a surprising number of sites fail without knowing it. Each AI company runs more than one bot, and they do different jobs.

For OpenAI, the split is documented clearly. OAI-SearchBot builds the index behind ChatGPT search, and OpenAI states that sites which block it will not be shown in ChatGPT search answers. GPTBot collects content that may be used for training future models. ChatGPT-User fetches a page when a person in ChatGPT asks it to, and is not an automatic crawler. These are independent, which means you can block training and still appear in search, if that is what you want.

Anthropic works the same way: ClaudeBot is the training crawler, Claude-SearchBot supports search results, and Claude-User fetches pages for a person's question. Perplexity uses PerplexityBot. Google has a token called Google-Extended that controls whether your content is used for its Gemini models, and blocking it does not remove you from Google Search.

If you want to be recommended, the search and user bots need to be allowed. A plain, permissive robots.txt does that already. If yours has grown a list of AI bot blocks copied from a forum thread years ago, read it again line by line.

User-agent: OAI-SearchBot
Allow: /

Then check the layer most people forget: your CDN or firewall. Several hosting and security providers now offer one-click AI bot blocking, and some switch it on by default for new sites. Your robots.txt can say "welcome" while your firewall returns a challenge page to every AI crawler.

Finally, register your site with Bing Webmaster Tools and submit your sitemap. OpenAI lists Microsoft among the search partners ChatGPT has used, and many practitioners believe Bing's index still feeds a good share of what ChatGPT finds. That is not confirmed in detail, but it takes twenty minutes and helps with Microsoft Copilot regardless.

Step 2: Make your pages readable without JavaScript

Here is a detail that catches out a lot of modern SaaS sites. Google renders JavaScript before it indexes a page. Most AI crawlers do not. Vercel looked at AI crawler traffic across its network and found that the major AI bots fetch JavaScript files but do not execute them.

If your homepage is a single-page app that arrives as an empty shell and builds itself in the browser, an AI crawler may see a title, a loading spinner, and nothing else. Your pricing and your carefully written comparison section simply do not exist as far as that bot is concerned.

The test takes thirty seconds. Open your pricing page, view the raw page source (not the inspector), and search for your actual prices. If they are there in plain HTML, you are fine. If the source is mostly script tags, look at server-side rendering or pre-rendering for your key pages. A browser extension like Siteprint captures the rendered version of a page as Markdown, which you can set beside the raw source to spot what only appears after scripts run.

You do not need to rebuild the whole app. Only the public pages matter here: homepage, pricing, features, comparisons, docs and the blog.

Step 3: Write the pages an assistant wants to quote

When an assistant searches on someone's behalf, it wants a passage it can lift into an answer, not your design or brand voice. The questions people ask about tools fall into a few patterns:

"What's the best tool for [job] for [type of user]?"
"[Your product] vs [competitor], which is better for [situation]?"
"Alternatives to [big competitor] that are cheaper or simpler"
"Does [product] integrate with [other tool]?"
"How much does [product] cost?"

Now ask whether your site has a page that answers each one in plain words. For most small SaaS sites, it does not. The homepage says "the modern way to manage your workflow," the pricing page hides numbers behind "contact us," and there is no comparison page because comparing yourself to a competitor felt awkward.

Lead with the plain answer

Put a one-sentence, literal description of what you do near the top of your homepage. "Help desk software for SaaS teams under 20 people, with a shared inbox, live chat and a help centre" is quotable. "Support, reimagined" is not, and the assistant will skip it for a competitor who said the boring thing clearly.

Use the same sentence, or close to it, everywhere you exist: your homepage, your social bios, your directory listings, your app store pages. Assistants piece together what a product is from many sources. When they all agree, you become a clear, well-defined thing. When every page describes you differently, you become fuzzy, and fuzzy products do not get recommended.

Publish prices as text

If an assistant cannot find your price, it will either guess, skip you, or quote an old figure from a third-party page. Put real numbers on the pricing page in plain HTML, with what each plan includes. If you genuinely sell custom enterprise deals, publish a starting price anyway.

Write honest comparison and alternatives pages

A page titled "[Your product] vs [competitor]" that admits where the competitor is stronger is far more likely to be quoted than one that claims you win at everything. Assistants are trying to give balanced answers, and a page that already reads as balanced is easy to use. Say who should pick the other tool. It costs you nothing with the people who were never going to buy, and it earns trust with the ones who might.

Say who it is not for

This sounds backwards, but "not a fit if you need on-premise hosting" helps an assistant match you to the right question, and it makes the rest of the page more believable. While you are at it, put visible dates on changelogs and docs. Freshness counts when an assistant picks between two similar pages.

Step 4: Get mentioned on the pages assistants read

This is the part people underestimate. You can have a perfect site and still not be recommended, because when the assistant searches "best help desk for small SaaS," it reads other people's pages, not yours.

Large studies of the sources ChatGPT, Perplexity and Google's AI features cite keep putting Reddit, YouTube, LinkedIn and Wikipedia near the top, with ChatGPT leaning on editorial roundups and review sites. The rankings shift between studies, but the lesson does not: third-party pages about products are a big part of what assistants read. So the work looks like old-fashioned PR, aimed more carefully.

Find the roundups that already rank. Search "best [category] tools" and "[competitor] alternatives" and look at page one. Those are the pages an assistant is likely to read too. Write to the authors with a genuine reason to include you, such as a feature nobody on their list has or a free plan. A short, specific note beats a template.

Show up in communities without spamming them. When people ask about your category on Reddit or in niche forums, answer properly, mention your product where it fits, and say you built it. One useful reply that stays up for years beats fifty promotional posts that get deleted.

Get listed where listings are actually written. Directories and review sites are third-party pages describing your product, which is exactly what an assistant is looking for. The catch is quality. A directory full of scraped, identical descriptions is a weak source, and those sites tend to drop out of search entirely, as we explained in our post on why duplicate descriptions make directories disappear. Pick directories where a person reads submissions and the listing has room for an original description. You can submit your tool to FeaturedLane for free. A person reads every listing before it goes live, which is the whole point of being here.

Make your mentions consistent. Every time someone writes about you, they repeat what they found. If your one-line description is clear and the same everywhere, that is the version that spreads.

What about llms.txt and schema markup?

You will see llms.txt recommended in almost every AI visibility checklist, so it is worth being straight about it. It is a proposed plain-text file that gives AI systems a summary of your site and links to your key pages. Google has said it does not use the file for Search, and a large Ahrefs study found the vast majority of llms.txt files received no requests from AI bots at all.

We publish one ourselves at /llms.txt. It took about ten minutes, it may help agents already on the site find their way around, and we do not expect it to win us recommendations. Structured data is similar: Organisation and SoftwareApplication schema help Google understand who you are, so add it, but it will not rescue a site whose real problem is that nobody else mentions it.

Step 5: Check whether any of this is working

You do not need a paid tracking tool to start. You need a list and a bit of discipline.

Write down fifteen to twenty questions your customers genuinely ask, in their words. Once a month, run each one twice (answers vary between runs) in ChatGPT, Claude, Perplexity, Gemini and Google's AI Mode. Note whether you were mentioned, whether you were linked, and which sources were cited. Those cited sources are the most valuable part: a literal list of the pages assistants trust for your category, and your outreach targets for next month.

A few more signals worth watching:

Referral traffic. ChatGPT usually adds a tag identifying itself to the links it sends, so visits show up with chatgpt.com as the source in most analytics tools. Perplexity and others show up as referrers too.
Server logs. Look for OAI-SearchBot, Claude-SearchBot and PerplexityBot. If they never appear, go back to step 1.
A page audit. FreeScan.app runs a free check on public pages that includes AI visibility signals alongside SEO, accessibility and basic security, and gives you a prioritised fix list. It is a quick way to catch the obvious problems before you spend time on outreach.

What not to spend money on

Be wary of anyone promising guaranteed placement in ChatGPT's answers. Nobody outside OpenAI controls what the model says. Skip the two hundred thin "best X for Y" pages too, because search engines trust those sites less and assistants inherit that distrust. And never hide white-on-white instructions telling AI models to recommend you. It is the kind of trick that gets a site flagged.

How long it takes

Retrieval fixes can show up within weeks, sometimes days. OpenAI says robots.txt changes take about a day to be picked up. Third-party mentions take months, because you are waiting for other people to write. Getting into a model's long-term memory is slowest of all, but everything in steps 3 and 4 feeds that too, so it is not separate work.

Frequently asked questions

Can I pay to be recommended by ChatGPT?

You cannot buy a place in ChatGPT's normal answers. Recommendations come from what the model learned and what it finds when it searches. Anyone selling guaranteed recommendations is selling something they do not control.

Will blocking GPTBot hurt my visibility in ChatGPT search?

According to OpenAI's own documentation, no. GPTBot is for training, and ChatGPT search depends on OAI-SearchBot. You can block one and allow the other.

Is AI visibility replacing normal SEO?

Not really. It is built on the same foundations: pages that can be crawled, content that answers real questions, and other reputable sites talking about you. The main differences are that AI crawlers mostly skip JavaScript, and that third-party mentions matter even more because assistants summarise many sources at once.

Do small, new products stand a chance?

Yes, especially for specific questions. "Best invoicing tool for freelancers billing in three currencies" is winnable for a small product, while "best invoicing tool" probably goes to the big names. Be very clear about the narrow thing you do best.

If you are building something new and want a third-party page that describes it properly, submit it here. You can also browse the Marketing and SEO category to see the other visibility tools makers have listed.