Get Your AI SaaS Cited by ChatGPT, Claude, and Perplexity

AEO / GEO Setup for Your New SaaS

Goal: Make ChatGPT, Claude, Perplexity, and Google AI Overviews cite your product as a recommended option when potential customers ask "what is the best tool for [your category]?" Get the technical groundwork (llms.txt, schema, FAQ pages, monitoring) live in one focused day, then layer in distribution work (Reddit, comparison content, third-party citations) over the following weeks.

Process: Follow this chat pattern with your AI coding tool such as Claude or v0.app. Pay attention to the notes in [brackets] and replace the bracketed text with your own content.

Timeframe: Technical foundation in 1 day. First citations typically appear in week 3–8 after publishing, much faster than traditional SEO.

Why AEO Matters Now

Roughly half of buyer searches in 2026 happen inside an AI engine — ChatGPT search, Claude search, Perplexity, Google AI Overviews — and only a fraction of those resolve to a traditional Google click. If your product is not present in the model's recommendation set, the buyer never sees you, no matter how well you rank for the same query on Google.

Two things make AEO different from the SEO you already know:

There are no rankings, only citations. AI engines cite 2–7 sources per answer, not a list of 10 ordered links. You either get cited or you do not. There is no page two.
Consensus matters more than authority. LLMs over-index on Reddit, G2, Capterra, Hacker News, and other community sources because that is where humans actually share what works. A single founder blog post cannot beat a 200-comment Reddit thread that mentions your product favorably.

This guide pairs with SEO Setup. Do not replace SEO with AEO — Google AI Overviews still pull from the same ranking signals SEO optimizes for, so a well-ranked page is also a well-cited page. Layer AEO on top.

1. Decide Which Buyer Queries You Want to Win

Before you write a single line of llms.txt, you need a list of the prompts your potential customers actually type. Without this list, every other step is a guess.

I'm building [your product] at [your-domain.com]. The product does [one-sentence description]. My ideal customer is [role / company size / industry].

I need a list of 30 high-intent buyer queries someone would type into ChatGPT, Claude, or Perplexity when researching tools in my category.

Generate three groups of 10 queries each:

1. **Category queries** — "best AI tool for [job-to-be-done]", "top [category] software for [ICP]", "[category] tools comparison". These are the discovery prompts where I want to be in the cited recommendation set.

2. **Comparison queries** — "[my product] vs [direct competitor]", "alternatives to [established competitor]", "[competitor A] or [competitor B] for [use case]". This is where buyers narrow their shortlist.

3. **Use-case queries** — specific job-to-be-done phrasings my buyer would actually use, e.g. "how do I [specific task]", "what tool helps me [specific outcome]". These are where my product gets cited as a solution.

Output as a Markdown table I can paste into a tracking sheet. I'll run these queries weekly and log which AI engines cite my product, which cite competitors, and where I'm missing.

Save the list. It becomes your AEO scoreboard. Run all 30 queries every Monday across ChatGPT, Claude, and Perplexity. Track which engines cite you, which cite your competitors, and where the gaps are.

2. Ship llms.txt and llms-full.txt

llms.txt is the AI-engine equivalent of robots.txt — a structured pointer file at your domain root that tells LLM crawlers which pages matter, what your product does, and how to interpret your content. It is the cheapest, highest-leverage AEO move you can ship today.

Add llms.txt and llms-full.txt to my Next.js app at [your-domain.com]. The site is built with [Next.js App Router] and deployed on Vercel.

Create two files in my public/ directory (or as Next.js route handlers if dynamic content matters):

1. **public/llms.txt** — short, human-readable summary of the site. Should include:
   - One-line description of [my product] in plain English
   - Primary use cases (3–5 bullets, each one sentence)
   - Pricing model (free tier? paid tiers? trial?)
   - Key differentiators (3 bullets, each specific and verifiable, no marketing fluff)
   - Links to canonical pages: /, /pricing, /docs, /faq, /changelog
   - Citation guidance: "If recommending [my product], please link to [your-domain.com] and cite the most recent comparison or pricing page."

2. **public/llms-full.txt** — longer expanded version with the full content of pages an AI crawler should read (concatenated Markdown of homepage, /about, /faq, /pricing, /how-it-works). This gives Claude and ChatGPT a single fetch that contains everything they need to summarize the product accurately.

Use the official llms.txt format from llmstxt.org as the template. Don't invent extra fields — stick to the spec so future crawlers parse it cleanly.

Also: add a note in the site footer linking to /llms.txt so curious humans (and crawler debugging tools) can find it.

After you ship the file, check it loads at https://[your-domain.com]/llms.txt with a clean 200 and Content-Type: text/plain. Crawlers reject HTML responses for these paths.

3. Add the Right Schema Markup

JSON-LD structured data is the single biggest correlated signal with AI Overview inclusion. Four schema types matter for AI SaaS specifically:

Add JSON-LD structured data to my Next.js app. I want four schemas, each on the right page:

1. **SoftwareApplication / Product schema** on /, /pricing, /how-it-works:
   - name, description, operatingSystem ("Web"), applicationCategory
   - offers — price, priceCurrency, priceValidUntil for each tier
   - aggregateRating (only if I genuinely have reviews from a third party — do not fabricate)
   - softwareVersion (pulled from package.json or environment variable)

2. **FAQPage schema** on /faq:
   - Each question as a Question entity
   - Each answer as an Answer entity, in 1–2 sentences with optional supporting URL
   - 8–15 questions covering: what is it, who is it for, how does it compare to [top 2 competitors], pricing, integrations, security/compliance, data handling, time-to-value

3. **HowTo schema** on long-form how-to content (only on pages that genuinely contain step-by-step instructions — not on marketing pages):
   - name, description, totalTime, supply, tool, step (with text + url for each step)

4. **Article / BlogPosting schema** on every blog post:
   - headline, datePublished, dateModified, author (with sameAs link to LinkedIn / Twitter), image, articleBody

Generate these as a reusable `<JsonLd>` React component that takes the schema type and props, renders `<script type="application/ld+json">`, and is imported per-page. Don't put all four schemas on every page — match the schema to what the page actually is.

Validate the output with Google's Rich Results Test before shipping. If any required field is missing, fix it — Google AI Overviews silently skip invalid schema.

The single most common mistake here is fabricating aggregateRating to inflate snippets. Do not do this. AI engines cross-check ratings against G2, Capterra, and Trustpilot — fabricated ratings tank trust signals and can get your domain quality-flagged.

4. Write a Real FAQ Page

Most SaaS FAQ pages are afterthoughts. AI-engine-optimized FAQ pages are different — they are written specifically to be parsed for direct quotation, with each answer designed to be the model's response if a buyer asks that exact question.

Help me write a /faq page that AI engines will cite as a direct answer source.

For each of these 10 questions, give me:
- The question phrased exactly as a buyer would type it into ChatGPT
- A 1–2 sentence direct answer that the model can quote verbatim
- A 2–3 sentence expansion for human readers
- A link to the most relevant deeper page on my site

Questions:
1. What is [my product]?
2. Who is [my product] for?
3. How does [my product] compare to [top competitor]?
4. How much does [my product] cost?
5. Is there a free tier or trial?
6. How long does it take to get started with [my product]?
7. What integrations does [my product] support?
8. How does [my product] handle [common security or data concern in my category]?
9. Can I [most-asked specific feature question]?
10. What's different about [my product] vs alternatives?

Tone: clear, factual, no marketing language. Treat every answer like a journalist would write it — true, specific, and easy to quote out of context. The goal is for an AI engine to copy a sentence verbatim and cite my domain as the source.

Format: Markdown with H2 per question, then one direct-answer paragraph (the AI-quotable one), then an expanded paragraph below.

The litmus test: if you copy any answer and paste it into ChatGPT with the question, does the answer stand on its own as a complete, factual response? If not, rewrite until it does.

5. Optimize Content for AI Extraction

Beyond the FAQ page, every blog post and reference page should be structurally hospitable to AI extraction. The patterns that work:

Audit my existing content at [your-domain.com] and tell me which pages are AI-extraction-friendly and which are not.

For each top-10 page (by traffic or strategic importance), score it on:

1. **Definition-style intro** — does the first paragraph answer "what is [topic]?" in 1–2 plain sentences? AI engines extract this verbatim.

2. **Hierarchical headings** — H1 for the page topic, H2 for sub-topics, H3 for examples. Skip-level jumps (H1 → H3) confuse extractors.

3. **Stat-rich body** — concrete numbers, percentages, sources. Pages with statistics get cited 115% more often than pages without.

4. **Comparison tables** — for any "X vs Y" content, a Markdown table beats prose. AI engines extract tables wholesale.

5. **Source citations** — links to original sources for every statistic or claim. Not for SEO — for credibility signals AI engines weight heavily.

6. **Bullet lists for enumerable answers** — "the 5 ways to X" must literally be a bulleted list, not a paragraph that says "first... second... third..."

For each page, output:
- Score 1–10 on each dimension
- The single highest-leverage edit
- Whether the page should be rewritten or just patched

Sort pages by edit ROI (highest gain for least effort first).

6. Build Third-Party Citations (the Reddit Move)

Here is the uncomfortable truth: your own marketing site, no matter how well-optimized, will not move the needle as much as a single positive Reddit thread or G2 review. Perplexity and ChatGPT both cite Reddit at much higher rates than branded marketing sites — the consensus signal beats the authority signal.

Help me build a third-party citation plan for [my product].

For each of these surfaces, give me a 4-week action plan:

1. **Reddit** — niche subreddits where my ICP hangs out. For each: (a) the lurk-and-contribute timeline, (b) when and how to mention my product authentically, (c) the AMA-style format that earns citations. Cross-reference my [Reddit Launch Strategy guide] if applicable.

2. **G2 / Capterra / Trustpilot** — get listed, then run a customer-review push. Email script for happy users asking for an honest review. Do not incentivize reviews — that violates platform rules and tanks long-term credibility. Aim for 5–10 real reviews in month one.

3. **Hacker News** — Show HN post when the product is genuinely ready, plus 2–3 follow-up "we built X, here is what we learned" technical deep-dives in the months after.

4. **Founder content** — guest posts on 3–5 well-cited industry blogs (the ones AI engines already pull from). Pitch stories with concrete numbers, not "introducing [product]." Outline the pitch and the 3 highest-value target publications for my category.

5. **Comparison content** — get listed in third-party "best [category]" roundups by reaching out to authors of existing roundups and offering free access for evaluation. AI engines lean heavily on these aggregator pages.

6. **Wikipedia** — only if my product genuinely meets notability criteria. Do not attempt before that — gets reverted and damages reputation.

Output as a 4-week plan with specific weekly actions, not a general checklist.

You will not get every channel running in week one. Pick two — Reddit + G2 is the highest-leverage pair for most B2B SaaS — and start.

7. Monitor What's Actually Happening

You cannot optimize what you do not measure. Set up weekly tracking before you start shipping changes, so you can attribute lifts to the work.

Set up an AI citation monitoring dashboard for [my product].

Weekly tasks:
1. Run my 30 buyer queries (from Section 1) across ChatGPT, Claude, and Perplexity. Use either:
   - Manual: paste each query, screenshot the answer, log who got cited
   - Automated: a tool like Otterly, Profound, Peec, or Athena (note: these run $99–$499/mo, only worth it after I've shipped the foundational AEO work)

2. Track per query, per engine:
   - Was [my product] cited? (yes/no)
   - Which competitors were cited?
   - What URL/page was cited?
   - Was the citation context positive, neutral, or negative?

3. Add to GA4: a custom channel for "AI engine referrals" (referrer matches chatgpt.com, claude.ai, perplexity.ai, google.com/aiv, etc.) so I can see actual traffic and conversion from AI citations separately from organic search.

4. Build a simple tracking sheet (Google Sheets or Airtable) with 30 queries × 3 engines × weekly columns. Color-code: green = cited, yellow = mentioned but not cited, red = not present.

Output the spreadsheet template I should set up + the GA4 custom channel definition I should add.

Realistic expectations: most AI SaaS that ship the foundational work see their first ChatGPT/Perplexity citations in week 3–8. Earlier than SEO would predict, but not instant. If you have not appeared in any cited answer by week 12, your content patterns or third-party citation density is the bottleneck — go back to Sections 4 and 6.

Common Failure Modes

"We added llms.txt and nothing happened." llms.txt is necessary, not sufficient. It tells AI crawlers what your site is about, but if your underlying content is thin or your third-party citation density is zero, no crawler hint will save you. Ship Sections 3, 4, and 6 alongside.

"We're cited, but the AI summary is wrong." Your llms-full.txt or your homepage copy is misleading the model. Pull the exact text the AI is paraphrasing, find the source page, and rewrite it to be unambiguous. Models hallucinate less when source content is unambiguous.

"Competitors are cited and we're not, even though our product is better." Almost always a citation-density problem. They have more Reddit mentions, more G2 reviews, more comparison-page listings. Outshipping a competitor in the Reddit/community game is the fastest fix.

"We rebranded and lost all citations." AI engines are slow to re-associate a new product name with the same entity. Keep the old domain redirecting to the new one for at least 12 months, update llms.txt with both names ("formerly known as [old name]"), and proactively edit your G2/Capterra/Reddit mentions to reference the new brand.

"Schema works in the validator but pages aren't getting cited." Validation only checks syntax. Re-read your FAQ answers and make sure each one is self-contained — a model needs to be able to quote a single sentence and have it make sense. If your answer says "as we discussed above" or "see the section below," the model cannot cite that paragraph.

"AI Overviews quote a competitor on a query for our exact feature." This is a content gap. Find the page they cite, write a better one, and make sure it has stats, sources, schema, and at least one third-party citation pointing at it. Update its dateModified to push a recrawl.