How to rank in ChatGPT search results: VISIBLE framework explained for B2B SaaS

A buyer at one of your best-fit accounts is close to choosing. They open ChatGPT and type a paragraph:

“We are a forty people team, have outgrown the spreadsheet we’ve been running, need something that handles a (specific workflow), and have to work with quickbooks. Please recommend software.”

ChatGPT gives them a few names. Yours isn't one of them.

That buyer is now evaluating your competitors. This is why how to rank in ChatGPT search results has become a live question for B2B SaaS teams, and you've probably run a version of that test yourself.

You typed the question your best customers ask, watched two or three competitors get named, and found your own brand missing from the answer.

The volume behind that gap is what makes it urgent. G2 surveyed 1,076 B2B decision-makers in March 2026 and found that 51% now start their software research in an AI chatbot more often than in Google, up from 29% a year earlier. ChatGPT is where most of that happens, and 69% of those buyers chose a different vendor than the one they walked in planning to buy, based on what the chatbot told them.

The part that stings is that you're probably not failing at SEO. You rank on Google for your category and your traffic chart looks reasonable. That's exactly why this loss stays hidden, because nothing in your reporting turns red when a model leaves you out of an answer.

I run this work for B2B SaaS clients through a framework I call VISIBLE, seven layers deep, and this article walks through all of them.

What ranking in ChatGPT search results actually means

There's no position three in ChatGPT. The model pulls together a set of sources it can reach, then writes one answer over them.

What you're chasing is two different outcomes, and it’s important to differentiate between the two.

• Recommended means your brand name appears in the answer.
• Cited means a URL from your domain was used as a source for it.

You can be recommended without ever being cited, and your blog post can be the source an answer is built on while the model recommends a competitor.

Recommended is what wins deals, because the buyer is reading names and building a shortlist from them. Cited is what feeds the model and what makes the case for naming you.

AI search strategy belongs to whoever already has SEO strategy right

The industry has spent two years selling GEO and AEO as a new discipline with its own budget line and its own specialist. Selling it that way costs money, because it hands the work to whoever is newest to the problem instead of whoever understands your buyers best.

Look at what each layer of AI visibility runs on. Ranking is what puts a page in the pool a model can retrieve from. The buyer research that makes content rank is the same research that makes a page match a buyer's contextual prompt.

Third-party mentions are the same relationship work that has always built authority, pointed at a new set of pages. Strip out the SEO foundation and the AI tactics have nothing underneath them. That's why brands who buy a visibility dashboard before they have anything worth measuring end up with a monthly chart and no movement.

Some of this genuinely is new, though, and I'd be overstating the case if I pretended otherwise. Prompt-level measurement didn't exist three years ago. The source-set problem has no clean Google equivalent.

Structuring a page so an answer can be lifted from the top of it matters more than it used to. All of that is real, and all of it sits on top of the older work rather than replacing it.

The results I can point to came from running it that way. With livepro, a knowledge management platform, top-ranking keywords went from 80 to 203 over twelve months, and more than 30 bottom-of-funnel keywords started appearing in AI Overviews. A field service management software client added $157K in monthly recurring revenue. An employee engagement SaaS grew demo bookings 2,388% year over year.

The framework I use is VISIBLE:

· Vet the buyer

· Identify high-intent keywords

· Substance over fluff

· Influence the sources

· Benchmark and track

· Live and fresh

· Entity clarity

You don't have to run them in that order. Vetting the buyer comes first because everything downstream depends on the language it produces, and skipping it makes the other six layers guesswork. After that, your entry point depends on where you already are.

V: Vet the buyer before you open a keyword tool

Every engagement I run starts with an onboarding form and a set of sales call recordings. Not a keyword tool. The form covers the product and its features, the competitors, the ICP, and any priority terms the client already has in mind. The sales calls are the input that matters most, and they're the one most teams never open.

Pull ten recent recordings and read them for the language your prospects are using. What you want is the way buyers describe the problem in their own words. Listen for the objections that keep coming up during evaluation and the phrases that show someone is ready to buy.

Then comes the point that makes this the layer everything else depends on. The way a buyer describes their problem on a sales call is close to the way they describe it to ChatGPT. It's the same person with the same problem reaching for the same vocabulary, a few weeks earlier in their process.

Couple that language with the keywords you're targeting and you have your prompt list. The words your buyers use, wrapped around the terms you're trying to win.

Keyword tools can't hand you this, because they index queries and your buyer isn't typing a query. They're typing a paragraph with context. There's no tool that stores that sentence, and no amount of search volume data reconstructs it.

I understand why teams skip this step. It takes a couple of weeks, it produces nothing publishable, and no dashboard moves while you're doing it. That cost is real. What it buys is every decision downstream getting made on real buyer language instead of an assumption.

In AI search that gap shows up faster than it used to. A model matching a buyer's situation to a page has nothing to work with when the page came from a keyword tool's suggestions.

Where possible after onboarding, I also interview internal stakeholders to understand the product properly. It's how the product knowledge gets deep enough for the Substance layer to have something real to work with.

I: Identify the high-intent keywords buyers search when they're ready to buy

I build the keyword strategy from the bottom of the funnel up. Ask ChatGPT which knowledge management tool suits a contact centre running on Confluence today, and it has to name products to be useful. These are also the prompts your dream clients run when they’re close to a buying decision, so it makes sense to exhaust these first.

The keyword families worth building around are the ones that force a named answer:

· "Best [category] for [industry]"

· "Best [category] for [use case]"

· "[Product A] vs [Product B]"

· "[Competitor] alternatives"

· "[Category] software for [vertical]"

· Role modifiers, like "[category] for [job title]"

A term with 30 monthly searches that hits search demand, product relevance and ICP pain-point alignment will outperform one with 3,000 that misses any of the three. I've written up the full method for finding bottom of funnel keywords separately.

The other thing this layer buys you is coverage. No single prompt can be won, because buyers phrase their situations in endless ways and each one produces a slightly different answer.

What you can do is cut at the category from many specific angles: by industry, by role, by growth stage, by integration requirement, by competitor comparison. Each angle catches a slice of the prompt space, and enough slices add up to a brand the model keeps encountering across the queries that matter.

Volume isn't meaningless, though, and I don't want to leave you thinking I'd cap a programme at 30-search terms forever. Bottom-of-funnel terms come first because that's where the pipeline is. Once those are covered, the strategy works up the funnel, and by then the domain has the authority to compete for the harder terms anyway.

S: Substance over fluff, because a model needs something specific to lift

Every piece has to clear two checks before it goes into production. Does it say something original about the topic. Would it hold value that an AI tool can't produce in seconds.

A model composing an answer is looking for a specific claim it can attribute to a source. Generic category copy gives it nothing to lift, so your page can sit in the retrievable pool and still contribute nothing to the answer that gets read. That's the cited-without-being-recommended problem from earlier, showing up as a content problem rather than a technical one.

The depth comes from the product itself. I review documentation, watch demo recordings or use the product, and pull in the client's team when a topic needs it. All of it goes into a brand knowledge base, a positioning and messaging document that every piece works from. That's what stops the writing drifting into industry-generic copy.

That research feeds writing with one job: say what the product does, who it's for, which use case it solves, and how it differs from the alternatives. That level of specificity matches your product to your dream clients when they ask LLMs like ChatGPT for product recommendations.

To clear the bar, I also enrich pieces past plain text with custom infographics, insights from podcasts and expert interviews, and findings from research and original data. Those are the elements that make a piece hard to copy.

I: Influence the sources ChatGPT already cites in your category

Ranking and content get you into the pool. What the other pages say about you decides what the model does with you. This layer is the off-page work, and it runs in two moves.

The first move is getting onto the pages the model already reads. Run your buyer prompts, look at which pages ChatGPT cites in the answers, and reach out to those pages. The ask is concrete: add the client's brand to the list. On a list of ten, the goal is a place in the top five, because position on the page shapes how much weight the mention carries.

But note that a model cites a limited set of pages when it answers any one prompt, so the winnable space is small by definition. You won't get onto every page. Some publishers ignore the outreach, some never reply, and some read the request and say no.

That ceiling is why the second move exists. Run keyword research for the keyword that ties back to the prompt, then go after mentions on the pages ranking at the top of that SERP. It's the same objective aimed at a much larger surface.

Backlinks belong in this layer too, pointed at the pages that need authority to rank for revenue-driving terms. As authority grows, harder keywords come into range, and each one opens more prompts you can show up for.

B: Benchmark and track what you get recommended for, not just cited for

The method I run is straightforward. Take 50 to 60 buyer-intent queries, phrased the way buyers phrase them rather than as keyword fragments, and run each one twice across OpenAI, Anthropic, Gemini and Perplexity. Running each query twice matters because models don't answer the same way every time, and a single appearance can be noise.

Two runs start telling you the difference between consistent visibility and a lucky roll. I've documented how I build this AI visibility audit if you want the mechanics. But you can use any AI visibility tool for this practice.

What comes out of it:

· Recommended: your brand named anywhere in the response

· Cited: a URL from your domain used as a source

· Share of voice: how often you appear against named competitors

· Blindspot queries: the prompts where you never showed up at all

Those first two signals move independently, and averaging them into one visibility score hides the problem you'd actually want to fix. A brand that's cited and never recommended has a positioning and brand problem, because the model trusts the page and still doesn't see a reason to name the company.

A brand that's recommended and never cited has a content problem, because the reputation is carrying it and its own pages aren't contributing.

The blindspot queries are the most useful output of the lot, since they tell you what to build next month without any interpretation needed.

L: Live and fresh, so pages get refreshed before the impressions slide

I watch two signals for this layer: rankings and impressions. When impressions drop on a page, and usually that's because rankings dropped first, the page goes on the refresh list.

The refresh starts with a fresh SERP analysis for the target query, because the reason for the slide is normally sitting in what changed around your page. Someone published something better, the intent shifted, or the format the query rewards has moved.

For a product-driven page, I also check whether the product itself has changed: new features, different pricing, anything the page asserts about how the product works. Those pages stay current as the product moves.

The consequence of skipping this is worse in AI search than it was in organic. A page making a stale claim about your product doesn't just lose a ranking. A model reading it will repeat that stale claim to a buyer who's comparing options right now, and you'll be answering for a price or a limitation you dropped two releases ago. Decay stopped being only a traffic problem and became a correctness problem.

E: Entity clarity, so every source tells the model the same story

If your site says you serve SMBs and three review platforms describe you as an enterprise solution, there's no consistent story for the model to repeat. So it hedges, or names a company whose positioning is clear across the internet.

Conflicting positioning confuses human readers exactly as much as it confuses models. A buyer who reads two contradictory descriptions of what your product is for doesn't investigate the discrepancy.

They discount you and move to an option they understand in one pass. Inconsistent positioning has always cost deals without anyone tracing the loss back to it, and AI answers made the cost visible by turning it into an absence you can measure.

The work is unglamorous. Find every place your brand is described: your own site, G2 and the other review platforms, industry directories, listicles, partner pages. Read what each one says about who your product is for and what it does. Where a source contradicts your positioning, get it corrected.

You won't get all of them, though. Some sites never respond, and some have no process for editing an old page. The realistic goal is consistency across the sources that carry weight in your category.

If you'd rather not build this alone, that's the work I do for B2B SaaS companies. Book a free strategy call and we'll map out what it would take to get you named in the answers your buyers are already reading.

Frequently asked questions about ranking in ChatGPT search results

How long does it take to show up in ChatGPT search results?

It depends on which layer is holding you back. Entity clarity fixes and listicle placements can land within weeks, because you're changing a source the model already reads. Content and authority work runs on a longer curve, closer to what you'd expect from organic rankings.

If I already rank on Google, will ChatGPT cite me?

Ranking helps, because it puts your page in the pool a model can retrieve from. It doesn't transfer on its own. The published studies on this disagree, and they disagree because they measure different things. Overlap between ChatGPT citations and Bing's top results is a different question from overlap between an AI answer's cited domains and Google's organic top ten. Treat your rankings as the entry ticket rather than the outcome, and measure your AI visibility separately.

Can a B2B SaaS with low domain authority compete in ChatGPT?

Yes, though authority still helps and I won't pretend otherwise. A narrow, specific prompt has fewer pages that genuinely match it, so a page written for that exact buyer situation can win against a much larger site that only covers the topic in general terms. Specificity is the lever you control, and on a narrow enough prompt it can outweigh the authority gap.

Do I need structured data or an llms.txt file to rank in ChatGPT?

Schema is worth having and it isn't the lever the industry treats it as. The statistic everyone repeats about cited pages using structured data comes from a single vendor study, and pages that get cited having schema is not the same as schema causing citations. llms.txt has thin adoption across the models that matter. Add both if they cost you an hour, and don't build a strategy on either.

How is ChatGPT visibility different from Perplexity or Google AI Overviews?

Each platform retrieves and cites differently, which is why I run audits across four of them instead of one. Perplexity leans harder on live search, and AI Overviews sit closer to Google's own index. What doesn't change is the strategy underneath. Buyer language, high-intent coverage, source influence and consistent positioning move all of them, so you're not building four separate programmes.

Author

Usama Khan

Founder, Usama Khan Consulting

Usama runs a boutique, revenue-focused SEO and AI search consultancy for B2B brands. He works with a capped number of clients each month, embedded as a senior fractional strategist. The goal is always the same: make organic a sustainable pipeline channel. When he’s not building search strategies, he’s probably watching cricket or learning more about coffee.

Follow for B2B marketing content