AI search optimization: a 7-step guide for SaaS teams

AI search optimization is the work of getting your product named and cited when buyers ask ChatGPT, Perplexity, Gemini or Google AI for a recommendation. Seven steps, in order: confirm crawler access, map the buyer questions, test for repeated answers, find the cited pages, get onto those pages, build your own answer pages, then re-measure.

Most of the numbers below come from a test run on 27 Sep 2026 that pulled 434 live AI answers from ChatGPT, Perplexity, Gemini and Google AI Mode, covering buyer questions across seven software categories (welaunch test of live AI answers, 27 Sep 2026, 06-aeo-live.md).

1. Verify that AI crawlers can access your content

Start here because every later step depends on it. Open your robots.txt and look for the bots that actually feed AI answers. Semrush lists OAI-SearchBot, GPTBot, ChatGPT-User, CCBot, ClaudeBot, Claude-User and Claude-SearchBot among the crawlers commonly blocked there. Many SaaS sites block them by inheritance, not by decision: a security plugin, a CDN rule, or a robots.txt copied from a template three years ago.

The second access problem is rendering. AI crawlers often skip JavaScript-heavy, client-rendered pages. If your pricing table or comparison page only exists after hydration, it may as well not exist.

How to check it worked: fetch your key pages with JavaScript disabled and confirm the text you care about is in the raw HTML. Then check your server logs for hits from those crawler user agents over a two-week window.

Common mistake: treating this as a one-time fix. A new firewall rule or bot-management setting can reintroduce the block months later, silently.

2. Map the buyer questions AI engines are actually answering

You are not optimizing for keywords. You are optimizing for the questions a buyer types in full sentences, like "best tool for a small team" or "alternatives to Y for a small team". Write out twenty to forty of these from sales calls, support tickets and your own churn interviews.

Two mechanics matter here. Google's guidance describes retrieval-augmented generation, also called grounding, which pulls fresh material through core Search ranking systems to answer a prompt. And query fan-out, where the model fires off a set of related queries of its own to gather more. So one buyer question can trigger several searches you never wrote down. Cover the sub-questions: pricing, integrations, limits, who it is not for.

How to check it worked: every question on your list should be one a real buyer asked out loud, and each should produce a meaningfully different answer when you run it.

Common mistake: writing prompts that contain your brand name. "Is [your product] good for teams?" tells you nothing. The questions that matter are the ones where you are not mentioned yet.

3. Test whether your brand is named consistently, not just once

Run each question at least four times per engine. The answers move. In the 434-answer test, only 9 of 32 brand appearances (28%) in ChatGPT answers were named in all four runs of the same prompt, and 6 of 21 (29%) in Perplexity (welaunch test of live AI answers, 27 Sep 2026, variance section).

Measured as set overlap, the brand lists ChatGPT produced for the same prompt across four runs matched at a mean Jaccard similarity of 0.54 across seven prompts that named any brand; Perplexity came in at 0.57 across five such prompts (same test, variance section).

How to check it worked: you have a count, not a screenshot. Per question, per engine: named in how many of four runs.

Common mistake: celebrating a single screenshot. Being named once in four runs is a 25% hit rate, and it is not the same thing as being the default answer.

4. Find the pages each AI engine is actually citing

This is the step that turns measurement into a to-do list. For each buyer question, collect the URLs cited in the answer. Those pages are your targets.

The engines do not behave alike:

Source: welaunch's test of 434 live AI answers to buyer questions in seven software categories, 27 Sep 2026, US.

Across the whole set, 106 of 434 answers (24%) cited reddit.com (welaunch test of live AI answers, 27 Sep 2026). Reddit is not a side channel. On Google AI Mode it appeared in 40% of answers (44 of 111).

Category changes the mix too. Among 63 answers to AI-visibility buyer questions, 12 cited reddit.com and 12 cited youtube.com; among 80 answers about choosing an AI visibility tool, 22 cited reddit.com (same test).

How to check it worked: a ranked list of domains by how often they appear across your questions, with the specific URLs under each.

Common mistake: optimizing for the engine you personally use. Google AI Mode cites few sources and leans hard on Reddit. Perplexity cites roughly three times as many per answer. The same work does not pay off equally on both.

5. Get your SaaS onto the pages that get cited

Now go down the ranked list. Three kinds of target, usually: listicles and review roundups, directories and listings, and community threads.

Directories are the least glamorous and the most tractable. welaunch's public directory database lists 273 launch and listing directories, of which 198 accept a free listing, 16 are free with a badge or backlink and 56 charge. By type it splits into 163 SaaS and startup directories, 95 AI tool directories and 15 agent and developer directories, with 100 accepting AI products only. Pick the ones that already show up in your citation list first.

Community threads need a different hand. On Reddit and Quora, write an actual answer to the question being asked, name your product once, and say you built it. welaunch handles this step by drafting the reply, listing or page, then having the founder review and submit it from their own account; on Reddit and Quora it names the product with disclosure and does not stuff in links.

How to check it worked: the page you targeted now contains your product, and it still appears in the citation list for that question a month later.

Common mistake: going after the biggest-name page instead of the cited page. A high-traffic roundup that no engine cites for your questions does nothing for this project.

6. Build your own pages that AI engines can quote directly

AI engines extract responses at the paragraph level. Write so a single paragraph can be lifted and still make sense: claim, number, qualifier. No paragraph that depends on the three above it.

Structure helps the parser. Use semantic HTML5 elements like article, section and footer, and add Article, Organization, FAQ and Author schema.

Substance matters more. Google's own guidance warns that commodity content like "7 Tips for First-Time Homebuyers" is typically built on common knowledge and adds little, and contrasts a first-hand review, which carries a perspective from personal experience, with a summary that restates what already exists. Your own usage data, your own benchmark, your own failure cases: that is the part no one else can restate.

One small mechanical point: AI crawlers will not click links, so name your sources in the text rather than hiding the attribution in an anchor.

How to check it worked: paste a paragraph of your page into an engine and ask what it says. If the answer is vague, the paragraph is not self-contained.

Common mistake: publishing twenty thin comparison pages at once. That is the fast route to the problem described in step 11 below.

7. Re-measure: track citation frequency over time, not a single snapshot

Set a cadence. Weekly is enough for most teams. Run the same question set, on the same engines, and log how many runs named you and which sources were cited.

Share of voice is the useful number. In the 434-answer test, Profound was named in 23 of 63 answers to AI-visibility buyer questions, with Otterly and Peec at 17 each (welaunch test of live AI answers, 27 Sep 2026). That shape, one leader and a cluster behind, is what you are trying to move.

This is the step teams skip, so it is worth automating. welaunch runs the founder's buyer questions on all four engines several times a week, keeps each answer in full, and tracks which pages the engines cite over time.

How to check it worked: you can plot your named-in-X-of-4 rate by month and by engine.

Common mistake: changing the question wording between runs. You lose the comparison.

What is AI search optimization, and how is it different from SEO?

AI search optimization is the practice of making your content frequently referenced and prominently featured by AI search systems like ChatGPT, AI Overviews and Perplexity. It targets mentions and citations rather than rankings, and goes by other names too: generative engine optimization (GEO), or SEO for AI search.

It is not a replacement for SEO. Google states that its generative features are rooted in core Search ranking and quality systems, which is why SEO best practices remain relevant, and ranking well in ordinary results improves your odds of being used as grounding content.

The reason to care: G2's 2025 Buyer Behavior Report found 29% of buyers now use AI-powered search tools more often than Google, Walker Sands reports almost 60% of searches ending without a click, and visitors arriving from non-Google AI systems like ChatGPT convert 4.4 times better than organic search visitors.

How content formats and recency affect what gets cited

Two things shape what gets pulled. First, format: extraction happens at the paragraph level, so a direct answer near the top of a section beats a long wind-up. Second, the content has to be worth quoting. A first-hand review gets picked over a restatement, and commodity tip-list content adds little.

Traditional rank is not the gate it is in classic search. Semrush reports ChatGPT cited pages ranking at position 21 or worse almost 90% of the time. A page that never cracks page one can still be the one quoted.

B2B vs B2C SaaS: does AI search optimization differ?

The mechanics are the same; the weighting changes. For B2B, third-party credibility does the heavy lifting: where backlinks were the SEO signal, mentions from reputable outlets, analysts and reviewers carry even more weight in AI search. Analyst coverage, review sites and category roundups are the pages worth chasing.

For both, write the way buyers speak. AI search optimization rewards natural language and context rather than keyword matching. And for B2B especially, the smaller volume is offset by quality: AI-referred visitors convert 4.4 times better than organic.

How long until you show up in AI answers?

Longer than a campaign, shorter than classic SEO in some cases. Because ChatGPT cites pages ranking 21 or worse nearly 90% of the time, you do not have to win page one first. But grounding still runs through search, so ranking well improves your odds, and that takes months.

Also plan for the surface itself moving. AI Overviews appeared on about 16% of searches as of November 2025, down from a peak near 25% in July. Volumes shift under you.

Common mistakes that waste the effort

Mass-producing near-identical pages. Google is explicit: creating separate content mainly to manipulate rankings or generative AI responses violates the scaled content abuse policy, and AI-assisted drafts still have to meet the Search Essentials and spam policies.

Assuming a link is a citation. Crawlers will not click through, so name sources in text.

Ignoring variance. With only 28% of ChatGPT brand appearances and 29% of Perplexity ones holding across all four runs (welaunch test, 27 Sep 2026), a single good answer is noise until you repeat the prompt.

FAQ

What is the 30% rule for AI?

There is no official rule by that name, but the number comes up in consistency testing. In a run of 434 live answers, 28% of ChatGPT brand appearances and 29% of Perplexity ones held across all four runs of the same prompt (welaunch test of live AI answers, 27 Sep 2026). Treat roughly a third as the share of mentions you can count on repeating, and test everything four times.

What is the 80/20 rule in SEO?

It is the Pareto principle applied to search: a small share of your pages and queries produces most of your results. In AI search optimization the same shape holds for cited sources. A handful of domains will account for most of the citations across your buyer questions, so rank them by frequency and work from the top.

Is SEO still worth it in 2026?

Yes. Google says its generative features are rooted in core Search ranking and quality systems, and ranking well improves your odds of being used as grounding content. The twist is that ChatGPT cited pages ranking 21 or worse nearly 90% of the time, so poor rank does not disqualify you.

What is AI search optimization called?

It also goes by generative engine optimization (GEO), or simply SEO for AI search. Some teams say AI visibility optimization or answer engine optimization. The work is the same: get named and cited in AI answers rather than ranked in blue links.

Are there tools to track AI visibility and citation frequency?

Yes, and the engines will name some for you. In 63 answers to AI-visibility buyer questions, Profound was named 23 times, with Otterly and Peec at 17 each (welaunch test of live AI answers, 27 Sep 2026). Whatever you pick, the minimum useful output is: how often each question named you, per engine, over time.

What is the ideal word count for content AI engines cite?

Word count is the wrong unit. Engines extract at the paragraph level, so the question is whether any single paragraph answers a question on its own. Write self-contained paragraphs with the claim and the number in them, and let the page be as long as the topic needs.

Sources

• How to optimize for AI search results in 2026
• AI Search Optimization: How to Make Your Brand AI-Visible | Walker Sands
• Google's Guide to Optimizing for Generative AI Features on Google Search | Google Search Central | Documentation | Google for Developers
• welaunch.sh
• welaunch test of live AI answers, 27 Sep 2026
• welaunch's test of 434 live AI answers to buyer questions in seven software categories, 27 Sep 2026, US

Related on welaunch

• SaaS marketing in 2026: the channels that still work for a small team
• AI Overviews tracker comparison 2026: how to see whether Google's AI names you
• The Founder's Referral Program Playbook: Turning Your First 100 Users Into a Launch Growth Engine