Getting cited and recommended · · By Gerald De Oleo
How to Implement AEO in Your GTM (Start at BOFU, Not TOFU)
The short answer: don’t try to cover the whole funnel. Start at BOFU. Take your product, ask who would use it and in what specific case, and turn that into the questions your buyer asks the model right before deciding. Those questions are your plan. Everything else is noise until you win those.
I sell this work, so read the rest with that in mind. But the reason to start at the bottom isn’t a preference — it’s where the data says the decision now happens.
The shortlist moved into a chat window
The old GTM map was clean. Awareness at the top, consideration in the middle, decision at the bottom. You pushed demand in from above and moved it down.
That map is breaking. G2 surveyed 1,076 B2B software buyers in March 2026. Half of them — 51% — now start software research in an AI chatbot more often than Google, up from 29% a year earlier. That’s not the number that should worry you, though. These two are:
- AI chatbots are now the single biggest influence on which vendors make a shortlist, at 54% — ahead of review sites (43%) and vendor websites (36%).
- 69% of buyers chose a different vendor than they’d originally planned, based on what a chatbot told them. One in three bought from a vendor they’d never heard of before the model named it.
Read those together and the implication is uncomfortable: the shortlist — the one thing your whole GTM is trying to get onto — is now assembled by a model, in about four seconds, often between two of your own touchpoints, and frequently before a sales rep ever hears the deal exists.
The decision question is no longer something you answer with your content. The model answers it. And it answers with what it read on pages that, mostly, aren’t yours.
Why BOFU, not awareness
Implementing AEO across your entire GTM at once is unpredictable. Too many questions, too many engines, too much you don’t control. Start at awareness — the broad category questions — and you’re fighting the incumbent on the ground where it wins, and you’ll wait months to see if anything moved.
BOFU is the opposite. Fewer questions, more specific, highest intent you’ll ever touch. Three reasons it’s the right place to start:
The budget is there. The buyer asking “what should I use for my case” has already decided to buy. The buyer asking “what is this category” hasn’t.
The traffic is pre-qualified. In Ruler Analytics’ 2026 benchmark — built on more than 110 million tracked sessions — AI-referred traffic converted at 5.8%, higher than organic search at 4.9%, because the model filters out the browsers before they ever click. You don’t inherit that pre-qualification on broad informational queries; you inherit it on decision ones.
It’s winnable. You will not win the broad category question — the incumbent owns it. You can own every specific one, and that’s where your ICP actually is.
So the approach is simple: win your niche’s decision questions first. Those you can win, and they’re the ones that move pipeline.
“If AI traffic converts so well, why are my numbers flat?”
Because converting well and appearing at all are different problems, and most B2B SaaS has the second one. Across independent audits, the large majority of B2B brands don’t surface when buyers query their category in AI tools. The conversion premium is real — but it only pays out if you’re in the answer in the first place. That’s what the rest of this is about.
Step 1: start from your product, not keywords
Don’t open a keyword tool. Open your product and ask two questions:
Who would use it? Not “B2B companies.” The real buyer: their role, team size, stage, stack, budget.
In what case would they reach for it? The specific moment your product is the answer. The problem that makes them search today, not in the abstract.
Each “who + case” combination is a segment of your ICP. And each segment asks the model a different question. That’s where you start — not from search volume, which barely exists for these questions anyway, but from the shape of your own buyer.
Step 2: turn that into your buyer’s questions
Take each “who + case” and write it the way your buyer would actually type it into ChatGPT or Perplexity, with the context attached.
| Keyword thinking | BOFU question thinking |
|---|---|
| “best attribution tool” | “what attribution tool for a 20-person B2B SaaS on HubSpot with a tight budget?” |
| “project management software” | “PM tool for a remote design team of 8 that needs client approvals?” |
| “data warehouse” | “cheapest warehouse for a seed-stage startup already using dbt?” |
The right-hand column is where the shortlist gets decided. Each one carries context and constraints, and the answer is a short list where you either are or aren’t. Make that list of questions. There are usually fewer than you’d think — somewhere between five and fifteen cover most of a B2B SaaS pipeline.
That list is your backlog. Not the one a volume tool hands you: the one that comes out of your own product and your own buyer.
Step 3: build the pages that decide the answer
The model answers those questions with what it read on third-party pages and on yours. Your job is to be one of the sources it trusts, in the right category.
The BOFU pages the model actually pulls from:
- Comparisons — you vs the competitor that keeps showing up in the shortlist
- Integrations — “works with [your ICP’s stack],” the constraint buyers attach most
- Segment pages — one per “who + case,” not one generic page for everyone
- Pricing and proof — the last thing a buyer validates before the call
One page per question, with the direct answer at the top — not three paragraphs in. Your blog explains things; these pages get you bought. More on the gap between showing up and being recommended in How to Get AI Recommendations (Not Just Citations).
“Isn’t one strong comparison page enough?”
No, because the model answers differently for each buyer context. “Best X” and “best X for a 10-person team on a tight budget” return different shortlists. A single generic page competes for the broad question you won’t win. A page per segment competes for the constrained questions you can. The work is in the specificity, not the volume.
A worked example, end to end
Say you sell a marketing attribution tool. Here’s the whole method on one segment.
Product: multi-touch attribution for B2B. Who: a demand-gen lead at a 20–50 person SaaS, running HubSpot, budget-conscious, burned by a tool that was too complex. Case: they just got asked by their CEO which channels actually drive pipeline, and their current setup can’t answer it.
The question that buyer asks the model: “what’s a simple attribution tool for a small B2B SaaS on HubSpot that won’t need a data engineer to run?”
What the model needs to recommend you: a page that answers exactly that — named integration with HubSpot, positioned for small teams, honest about what it doesn’t do (enterprise complexity), with proof. Not your homepage. Not “the #1 attribution platform.” The specific answer to the specific constraint.
How you know it worked: you run that exact question across ChatGPT, Perplexity, Gemini and Google once a month. Either you’re in the shortlist, described as the simple HubSpot-friendly option — or you’re not, and that page is the next thing you build or fix.
Why this one matters for attribution specifically: these are the buyers the model is most likely to hand to a competitor, because the category’s buyer’s guides were written by the incumbents. I go deeper on that dynamic in Why ChatGPT Recommends Your Competitor’s Attribution Tool.
Step 4: check they can read you, and measure monthly
Two things before you call any of this done.
First, that AI can read the pages at all. Most AI crawlers don’t run JavaScript. If your pages don’t render their content without JS, they don’t exist to the model — and no amount of good writing fixes a page the crawler never sees. Check it in two minutes with the crawler check: paste a URL, see what each bot gets.
Second, that you’re described right when you do appear. Making the shortlist isn’t the finish line. You can be cited accurately in the category you moved away from two years ago — present in the answer, described as a company you no longer are. That’s why measurement is per engine and per question, every month. ChatGPT, Perplexity, Gemini and Google draw from different sources, disagree often, and change constantly. What worked in March stops working.
“How often does the ground actually shift?”
Enough that a one-time audit expires fast. The engines re-weight sources, competitors publish, and your own positioning moves. The teams that treat this as a monthly loop — same questions, every engine — are the ones who catch a slipped description before it costs a quarter. The ones who audit once and file it away are optimizing for a snapshot that’s already stale.
Quick answers
Where do I start an AEO strategy for GTM? At BOFU. Take your product, define who uses it and in what case, and build your buyer’s decision questions. Those are your plan — not broad awareness terms you won’t win.
Why BOFU and not top of funnel? Because that’s where the shortlist now forms — 54% of B2B shortlists are shaped by AI chatbots (G2, 2026) — and where intent, budget and winnability are highest. AI-referred traffic also converts higher than organic (5.8% vs 4.9%, Ruler Analytics 2026) precisely because it’s decision-stage.
How many questions do I need? Few. Five to fifteen BOFU questions, with your ICP’s context attached, cover most of a B2B SaaS pipeline. You don’t need a hundred.
Blog or service pages? Both, but don’t confuse them. The blog explains; the comparison, integration, segment and pricing pages are what the model uses to recommend you.
How do I know it’s working? Run your BOFU questions across every engine, every month. See whether you’re in the shortlist and how you’re described. If you’re not there, that page is the next one you build.
How we work
The four steps above are the order I run with clients, and it starts at the bottom on purpose.
First, I take your product and your ICP and build the real decision questions your buyer asks — with their stack, team size and budget attached. Before we talk, I run three or four of them in one engine — a free Shortlist Check — so the first call opens with something real instead of a pitch. The full run across ChatGPT, Perplexity, Gemini and Google is the audit: where you sit in the shortlist, how you’re described, who takes your place when you’re not.
From there the work is ordered — fix what describes you wrong before publishing anything, build the pages that answer the constrained questions, then measure the same questions every month. Full scope is on the services page.
Want to see what AI answers to your buyer’s real decision questions right now? Send me your domain and I’ll run a free Shortlist Check: three or four of your BOFU questions in one engine — whether you appear and how you’re described. The full run across every engine is the $499 audit. Book a call.