By
Austin Coker
GEO for SaaS is the work of making sure AI assistants have accurate, current, well-sourced facts about your product on every surface they draw from. When a buyer asks for a recommendation, your product should show up and be described correctly.
Most of that work happens off your website. When ChatGPT, Perplexity, Claude, Gemini, or Google AI Mode answers “best CRM for a 30-person sales team,” it builds the answer from:
Controlling what those sources say about you is the entire job.
This moved up the priority list because buyers moved first. G2 surveyed 1,076 B2B software buyers in March 2026 and found that:
Gartner’s March 2026 survey of 646 B2B buyers found that 67% prefer to buy without talking to a sales rep, and 45% used AI during a recent purchase. Buyers are building shortlists before they ever talk to you, and an AI assistant is describing your product to them.
Almost every software question buyers put to an AI assistant falls into one of five patterns:
An AI assistant doesn’t look up a software recommendation. It builds one.
Step 1: The question gets broken apart. Google documents that its AI features break a question into multiple related searches across subtopics and data sources before writing a response. It calls this “query fan-out.” A prompt like “best help desk software for a 50-person support team with Salesforce” quietly becomes separate searches about:
Other assistants follow a similar pattern.
Step 2: The assistant pulls passages, not pages. The units it grabs are small: a single paragraph from an integration doc, one row of a pricing table, one sentence from a review that mentions team size. Your beautifully argued homepage narrative doesn’t answer anything specific, so it rarely gets used. A line like “the Salesforce integration syncs contacts and cases both ways, every 15 minutes on the Growth plan” does.
Step 3: The assistant reconciles conflicts and writes the answer. This is where the damage happens. If your pricing page says one thing and a two-year-old review profile says another, the assistant often picks whichever is stated more plainly or repeated more often.
It then writes a short label for each product it lists, such as “best for enterprise teams with compliance requirements” or “good for startups, limited reporting.” You didn’t write that label. It reflects how third-party sources describe you. If review titles, roundups, and community threads keep calling you the cheap, lightweight option, that’s the label you get, whatever your positioning deck says.
Research shows this can be influenced. The paper that named the field, Aggarwal et al., “GEO: Generative Engine Optimization,” was presented at KDD 2024, a major data science conference. The researchers tested content changes against a benchmark of AI-generated answers. They found visibility gains of up to 40%, with results varying significantly by subject area.
Treat that as a controlled research result, not a business forecast. The useful takeaway for SaaS is the second finding: because results vary by subject area, generic advice underperforms. The work has to be shaped around how buyers evaluate your specific category.
This prompt family produces the shortlist, and it’s the one most SaaS companies lose without knowing it.
An assistant can only include you in a category it believes you belong to. Say your site calls you a “revenue intelligence platform,” but every review profile, directory, and roundup files you under “sales engagement.” You’ll show up for one and disappear from the other.
Pick the category name buyers actually use in their prompts, and use it consistently across your site, your review profiles, and every directory listing you control. Inventing a new category can be a growth strategy, but it isn’t a GEO strategy. It costs you visibility until the market adopts the term.
Almost nobody asks for “best project management software” and stops there. They add a team size, an industry, the tools they already use, a budget, or a workflow. Each of those qualifiers leads to a separate search behind the scenes.
A page built around your category plus one qualifier gives the assistant a passage to pull, as long as the qualifier shows up in the body text and not just the title. In our client work, the fastest-moving pages are a small set of segment and use-case pages that:
This feels counterintuitive to most founders. It works because the assistant is effectively deciding whether you’re a good fit.
A page that says “if you need multi-entity accounting or have fewer than ten seats, this isn’t the right tool” gives the model a clear signal about who you don’t serve. That makes your positive claims more believable. Vendors who claim to be great for everyone get described in vague terms, and vague descriptions lose shortlist spots to specific ones.
Review platforms are the other half of this prompt family. The star rating matters less than the review text. The passages that support “best for X” answers are reviews that mention:
When you run review campaigns, ask customers to describe their situation, not just to praise you. One review that says “we’re a 40-person agency and replaced a spreadsheet plus two tools” does more for GEO than fifty that say “great product, great support.”
These two prompt families are where buyers with budget and intent show up. Someone asking for alternatives to a named product has already chosen the category and is now choosing the vendor.
The alternatives page does most of the work here, and most SaaS companies build it badly. A page that calls the incumbent expensive and bloated, then lists four products you happen to beat, reads as marketing to an AI model just as it does to a person. Assistants discount one-sided pages because the rest of what they read contradicts them.
Instead:
What gets pulled from comparison content is concrete:
Marketing adjectives get ignored. If your comparison page can’t be fact-checked line by line against the competitor’s public docs, it won’t be trusted or used.
Buyers asking about alternatives immediately want to know what switching will cost them. A migration guide answers that and gives assistants a strong passage to pull. It should cover:
Content explaining why teams switch works too. Cover the specific reasons teams leave a given incumbent, described honestly rather than as a hit piece.
You can’t control a competitor’s comparison page. You can, however, correct what third-party comparison sites, pricing aggregators, and forum threads say about you. Those are often the passages an assistant reaches for when the two vendors’ pages disagree.
Pricing is one of the most common factual errors we find when auditing what AI assistants say about a SaaS product. It’s also one of the most damaging. A wrong price either rules you out before a conversation or sets an expectation your sales team has to walk back.
The errors come from predictable places:
The biggest single cause is a pricing page with no numbers. You haven’t kept your pricing private. You’ve left a gap, and the assistant fills it with whatever third-party estimates exist, which are usually wrong.
For sales-led SaaS, this is a trade-off to make deliberately, and buyer preference data points one way. When two-thirds of B2B buyers prefer to buy without a rep, a pricing page with no numbers screens out buyers who research on their own, at exactly the moment they build their shortlist.
At a minimum:
If enterprise pricing genuinely varies, publish a starting point and the factors that change it. A stated range always beats a number the assistant made up.
Google documents SoftwareApplication structured data, code that describes a software product to search engines, including its price. It won’t get you into AI answers on its own. Google’s AI features documentation is clear that no special markup is required to appear. What it does is state your price in a format machines read without ambiguity, so your number wins when an assistant is reconciling conflicting sources.
Once your own page is fixed, correct the copies elsewhere: review platform profiles, marketplace listings, directory entries, and pricing aggregator pages. Nearly all of them have a way to request updates, usually free, and most vendors never use it. It’s unglamorous work, but it removes more misinformation per hour than any content project.
Integration questions are quietly the highest-leverage prompt family in B2B SaaS because they can knock you out. A buyer asking “does this work with HubSpot and Snowflake?” isn’t comparing philosophies. They’re checking whether you make it to the next round. If the assistant can’t confirm the integration, it either says it’s unsure or leaves you out, and the buyer moves on.
A single “Integrations” page with a wall of logos is close to useless here, because a logo doesn’t answer the question. Each integration page should cover:
Get listed on the other side too. The app marketplaces inside the platforms your buyers already use, such as major CRMs, data warehouses, communication tools, and billing systems, are independent sources with a lot of authority on exactly this question. They’re often the first place an assistant checks to confirm an integration exists, because the platform vendor is a neutral party.
AI crawlers can’t read integration documentation that sits behind a login, inside a JavaScript app that shows crawlers nothing, or in a PDF, so for GEO purposes it doesn’t exist. This is the most common technical failure we fix on SaaS sites. The marketing team usually doesn’t notice, because engineering owns the docs.
Run this before commissioning any content. It’s manual and takes a day or two, and it consistently turns up errors that are costing deals right now.
Product documentation is the most underused GEO asset in SaaS. It’s the most factual content your company produces, and it’s written in the plain, direct style assistants favor. It also answers the narrow questions buyers ask late in an evaluation. Most companies leave it out of the marketing conversation entirely.
Changelogs matter because they’re how assistants learn a feature now exists. If you launched single sign-on last quarter and assistants still tell buyers you don’t have it, you need three things:
Undated changelogs, or release notes that only exist inside the product, don’t help.
Help center content covers buyers’ final objections: data export, permissions, data retention, uptime and support commitments, and onboarding. Buyers ask AI assistants these questions constantly, and the answers come from whatever is public.
This section is short because the requirements are narrow, but getting it wrong makes everything else pointless.
OpenAI. OpenAI says that any public website can appear in ChatGPT search as long as it doesn’t block OAI-SearchBot, its search crawler. OAI-SearchBot and GPTBot are separate crawlers with separate purposes: GPTBot collects content for training OpenAI’s models.
Plenty of SaaS companies blocked every AI crawler in 2024 as a protective default and never revisited it. That now quietly keeps them out of AI search results they want to be in. Check your robots.txt file and also your CDN’s bot rules, because the block is often set at the CDN rather than in robots.txt.
Google. For AI features, Google’s requirements are that pages be indexable and eligible to show a text snippet, with no special markup needed. Check that you aren’t suppressing snippets across the whole site with a nosnippet tag or a restrictive max-snippet setting. Some SaaS sites inherited these from an old content-protection policy.
The practical rule for SaaS: never put pricing, docs, integration pages, or comparison pages behind a login, an email form, or a JavaScript setup that shows crawlers an empty page. If a page matters to the buying decision, anyone should be able to read it without an account.
Measure product-led and sales-led motions separately. Mixing them produces numbers nobody trusts.
Track self-serve signups, trial starts, and activation from AI referral traffic. GA4 added an AI Assistant channel on May 13, 2026, which labels traffic from recognized AI assistants with the medium “ai-assistant.”
Use it, but treat the number as a minimum rather than a full count. A meaningful share of AI-influenced traffic arrives as direct traffic, because people read a recommendation and then type your domain. Watch the quality of that channel’s signups too. In our client accounts, AI-referred signups tend to arrive further along in their evaluation.
Track demo requests, opportunity source, and the pipeline each source produces. Add AI assistants as options in your “how did you hear about us” field. List ChatGPT, Perplexity, Claude, Gemini, and Google AI Mode separately rather than as one “AI” option, because the split tells you where to focus.
What buyers report about themselves is imperfect, but it’s still the best signal available for this channel. It also gives your sales team a visible reason to care about the audit work.
When demos and rankings are the goal, the results look like the B2B financial data platform that got 40% more demo signups and 3.3x more top-three rankings within five months. Another example is an earlier-stage construction SaaS company that went from zero to more than 2,000 monthly visitors, with its domain rating rising from 0 to 31.
For tying revenue directly to AI search, the clearest example we have is a staffing firm that tracked $70K in revenue from ChatGPT within four months. The same attribution method works for SaaS.
Re-run your audit prompts monthly. Record whether you appear, in what position, and with what label, and do the same for your main competitors. This is the closest thing to rank tracking for AI search. The label matters as much as whether you appear at all, since being included but described badly is its own problem.
Confirm that OAI-SearchBot and Google’s crawlers are actually fetching your pricing, docs, and integration pages, and how recently. Crawler visits confirm that the assistants can reach your content and how fresh their copy is. They don’t confirm that you’re cited, included in answers, or influencing buyers, so don’t report them to your board as a performance metric.
For detail on ChatGPT specifically, see our B2B SaaS ChatGPT guide, or read more about our SaaS search marketing.
Austin Coker is the CEO and founder of 95 Projects. He has worked in SEO since 2020, when he built and sold an ecommerce brand that grew mainly through search. He founded 95 Projects in 2022 to help SaaS, ecommerce, and B2B service companies show up wherever their buyers search, from Google to ChatGPT and AI Overviews.
Have questions about working with us? Book a 30-minute strategy call to discuss your goals and see if we’re a good fit.
No-pressure conversation
Clear next steps (if we’re a fit)
Talk directly to an expert
GEO for SaaS is generative engine optimization applied to software: getting your product accurately represented and recommended in answers from ChatGPT, Perplexity, Claude, Gemini, and Google AI Mode. It covers your own pages and the third-party sources assistants draw from, including review platforms, integration directories, marketplaces, docs, and community threads. The practical unit of work is the prompt family rather than the keyword.
SEO is about ranking in a list of pages, so success means a high position. GEO is about AI-written answers, so success means being included in the recommendation and described accurately. The two overlap heavily, since both require pages that search engines can index and show snippets from. GEO puts much more weight on consistent facts across sites you don’t own, and on pages traditional SEO programs often ignore: pricing, integrations, and documentation.
Correcting facts that already exist, such as a wrong price or a missing integration, tends to show up fastest once the source is fixed, because the assistant is resolving a conflict rather than discovering something new. Getting into a competitive “best [category] software” answer with little review presence takes much longer, because it depends on third-party coverage building up. Expect the first audit to improve accuracy quickly, and plan on several quarters of work to get onto shortlists.
Yes, more than most SaaS teams assume. Review platforms are neutral third-party sources with structured information on category, pricing, and customer sentiment. G2’s research found that AI tools are now the top influence on shortlists and that 51% of buyers start research in AI tools more often than in Google. Review content is part of what those tools read. Focus on reviews that mention company size, industry, the tools the customer uses, and the product they switched from.
In almost every case, yes: at least plan names, a starting price or range, and how seat count or usage drives cost. A pricing page with no numbers doesn’t keep your pricing private. It leaves the question to outdated review profiles, old blog posts, and aggregator estimates, which are usually wrong. If pricing genuinely varies by deployment, publish a starting point and the factors that change it.
It helps make your facts unambiguous, but it won’t get you included on its own. Google documents SoftwareApplication structured data that includes price, while its AI features documentation says no special markup is required to appear in AI experiences. Use it so your price and plan details exist in a clear, machine-readable form, and don’t expect it to replace correcting the third-party sites that carry old numbers.
