How to get your SaaS recommended by ChatGPT and Perplexity

From the SoleOS answers series — written about our own product space; grounded in published definitions and documented behavior, never invented numbers.

Getting recommended by ChatGPT, Perplexity, or any AI assistant comes down to three things: your product's information has to exist in a form a model can extract cleanly, it has to show up on sources those systems already trust, and it has to answer the specific question a buyer is typing rather than a generic pitch about your category. There is no confirmed ranking algorithm to reverse-engineer here — these tools blend training data with live retrieval (Perplexity and ChatGPT's browsing mode search the web at query time; other assistants lean more on what was in their training corpus), so the honest goal is to make your product easy to find, easy to quote accurately, and easy to trust, then measure what happens next.

Disclosure: SoleOS — a portfolio dashboard for solo founders — publishes this guide because AI-search visibility is part of the distribution problem our own users deal with. We've tried to keep it implementation-focused rather than a pitch for the product.

Why AI assistants recommend some products and not others

Two mechanics are at play, and it matters which one you're optimizing for. Pure-training-data answers (a model recalling what it learned, with no live search) depend on how often and clearly your product was described in text the model ingested — documentation, comparison articles, forum threads, review sites. Retrieval-augmented answers (Perplexity's default mode, ChatGPT with browsing/search on, and increasingly most assistants) run a live web search, fetch a handful of pages, and summarize them on the spot. That second mode is closer to classic SEO than people assume: a page that ranks, loads fast, and answers the query directly can get pulled into an answer today, without waiting for a model retrain. Both mechanics reward the same things — clarity, accuracy, and presence on pages that already rank — which is why most of the tactics below help regular search too.

Answer the exact question, not the category

Nobody asks an assistant "what's a good SaaS tool." They ask something specific: "what's the best way to track MRR across multiple RevenueCat apps," or "how do I see App Store and Stripe revenue in one dashboard without building it myself." Write for that phrasing. A page titled and structured around "how to track revenue across multiple apps" will get pulled into an answer for that query far more often than a homepage that talks about "unifying your business metrics." This is long-tail intent: the more precisely a page matches the words and structure of a real question, the more extractable it is. If you sell to a narrow audience (solo founders running several products, in SoleOS's case), your best AI-search content is the specific workflows that audience searches for — see our guides for the pattern: pages built around one concrete founder question rather than a category pitch.

Structure content so a model can lift the answer out

Assistants extract answers, they don't read essays. Three structural habits make extraction easier: open every important page or section with a direct, self-contained answer in the first sentence or two — no throat-clearing; use real headers phrased as questions where that fits, since both search engines and models weight headers heavily; and include an actual FAQ block with short, complete answers, because FAQ-formatted content maps almost one-to-one onto how these tools phrase their own responses. Define your terms plainly too — if you use a term of art (like "portfolio intelligence" or a specific integration name), state what it means in one sentence near first use rather than assuming context. None of this requires new content; it's often a rewrite of an intro paragraph and the addition of a definitions sentence.

Publish an llms.txt file — and keep it honest

llms.txt is an emerging, informally-adopted convention: a plain-text file at your site's root describing what the product is, who it's for, and linking to your key pages (docs, pricing, a demo) in a format built for machines to parse rather than humans to browse. Support is inconsistent — some crawlers and agent tools read it, plenty don't yet, and no assistant is guaranteed to use it. Publish one anyway; it costs little and can only help as adoption grows. The one rule that matters: write it like an honest one-paragraph pitch to a person, not marketing copy. If it claims things your actual pages or reviews contradict, you've handed a model a document that makes your own site look inconsistent the moment it cross-checks.

Be worth citing in the first place

The tactics above only work on content that's actually good. Model providers are visibly tightening what their pipelines treat as authoritative, and thin, templated, or reworded-from-a-competitor content is exactly what those filters are built to catch. The durable version of "AI SEO" matches the durable version of regular SEO: publish something original — a workflow you actually use, a trade-off you had to make, numbers from your own product rather than vague industry averages — and be accurate about your own limitations. A guide that says "here's when you don't need this tool" is more citable, not less, because it reads as trustworthy rather than promotional.

Earn mentions on sites the models already trust

Both training-data recall and live retrieval favor products that show up on independent sources: comparison and "best of" articles, Reddit and Hacker News threads where real users discuss the category, review and directory sites, podcast show notes, and other people's blogs that link to you. You can't buy your way onto most of these credibly, but you can earn them the normal way — be genuinely useful in communities where your buyers already are, give reviewers and comparison-article writers something specific and true to say about you, and make it easy for anyone writing about your space to link to a concrete page — an integrations list, a pricing page — instead of just your homepage. A mention on a site an assistant already treats as a credible source is worth more than a dozen pages on your own domain.

Measure what's working, knowing it's still early

Track AI referral traffic the same way you'd track any other channel: segment sessions by referrer domain (chatgpt.com, perplexity.ai, claude.ai, and similar) in whatever analytics you already run, and watch the trend rather than the absolute number — it's small for almost everyone right now. Branded search volume (people searching your product's name directly in Google) is a decent secondary signal, since AI-assistant discovery often shows up later as someone searching your name to verify what they were just told. Running several products makes this worth wiring into a shared metrics view rather than checking analytics per app — see how to track analytics as a solo founder for the broader setup. Be honest about attribution here: this channel is new, volumes are low, methodologies vary by assistant, and much of it is genuinely outside your control — no one, including the assistants' own makers, has published a reliable way to audit why a given answer cited what it cited.

When you don't need any of this

If you run one product with a handful of customers and you already know how they find you, AI-search optimization is a low-priority afternoon project, not a strategy — put the time into the channels you can already measure precisely. It also matters less if your buyers are enterprise procurement teams working from an RFP rather than consumers or indie buyers typing questions into a chat window; that audience isn't asking assistants for recommendations yet in most categories.

Frequently asked questions

Does llms.txt actually improve my chances of being recommended?

Nobody outside the model providers can confirm how much weight it carries today, and support across tools is inconsistent. It's cheap to publish, aligned with good practice (a clear, honest description of your product), and can only help as more crawlers and agents start reading it — treat it as a hedge, not a guaranteed lever.

Should I optimize for ChatGPT specifically, or all assistants at once?

Optimize for the underlying behavior, not one product. Clear first-paragraph answers, FAQ blocks, accurate content, and external mentions on trusted sites help with ChatGPT's browsing mode, Perplexity, Claude, and traditional search simultaneously, since all of them are pulling from the same pool of well-structured, credible pages.

How long does this take to show results?

There's no fixed timeline, and it varies by mechanism — live-retrieval tools like Perplexity can surface a newly published, well-ranked page within weeks, while training-data recall only updates when a model is retrained, which can take much longer and isn't something you can trigger yourself. Expect this to compound slowly alongside your regular SEO and content work rather than produce a fast, isolated result.

Can I actually track when someone finds me through an AI assistant?

Partially. Referrer-based tracking catches sessions where the assistant includes a clickable link and the user clicks it, which undercounts real influence — plenty of people get a recommendation, then search your name directly or type your URL from memory, which shows up as direct or branded-search traffic instead. Track both signals together rather than relying on referrer data alone.

Does this replace regular SEO?

No — it's the same foundation applied to a new distribution surface. Pages that already rank well and answer questions directly are the ones most likely to get pulled into an AI-generated answer too, so investment in one reinforces the other.