Someone sat down, opened ChatGPT, and typed: “best operations consultant for a founder-led agency that’s scaling too fast.” Three names came back with a short reason next to each. They picked one, booked a call, and signed. They never searched your site. They never saw your LinkedIn. They never knew you existed.
You weren’t in the room. There was no pitch you lost. There was a shortlist you weren’t on, and no way of knowing it happened. No bounce, no abandoned cart, no “we went with someone else” email. Just silence, which you probably read as a slow month.
Most people treat AI visibility as a single switch: either AI knows you or it doesn’t. It’s three separate things, and you can be strong at one while completely absent from another. The cruel bit is that the layer most service businesses are best at is the one that matters least, and the layer that decides who gets the call is the one they’re invisible at.
What are the three layers of AI visibility?
Think about how someone actually buys when they use an AI engine instead of Google. They move through three stages, and each one is a different kind of question.
Layer one, informational. The potential client is learning the category. They don’t know you, and at this point they don’t need to. The prompts look like “what is answer engine optimisation” or “what tools do agencies use for ops.” They’re orienting. They want to understand the space before they shortlist anyone in it.
Layer two, consideration. Now they’re comparing. They’re building a shortlist for their specific situation. The prompts get sharper: “best ops consultant for a service business doing two million,” “alternatives to [the one name they’ve heard of],” “X versus Y for a small team.” This is where the decision narrows from everyone to three.
Layer three, conversion. They have a name, yours or a competitor’s, and they’re checking it. “Is [your business] any good.” “[You] versus [competitor].” They’re not discovering you here. They’re confirming a decision they’ve half made.
Three different questions, asked at three different points in someone’s head. Your visibility at each one is built by different content doing a different job. Being cleanly citeable when someone asks “what is X” has almost nothing to do with being named when someone asks “best X for my situation.”
The consideration layer is where you go dark
Almost everyone runs the conversion test first, because it’s the most flattering one. They ask ChatGPT “what does [my business] do” or “is [my business] good,” and the engine gives a warm, accurate answer. Relief. “AI knows me. We’re fine.”
They are not fine. That test only proves the engine can describe you once you’ve handed it your name. Of course it can. Your site says what you do. Your LinkedIn says what you do. Once the engine has the entity, it can read the file.
The layer that decides who gets the call is consideration, and that’s where most service businesses go dark, for a structural reason: their content describes what they do but never positions them against an alternative or a situation.
To build a shortlist, the engine needs to know who you beat, for whom, and when. “Best X for a founder-led agency.” “X versus the obvious incumbent.” “Alternatives to the tool everyone defaults to.” If nothing you’ve published answers a comparison or names a specific client situation, the engine has nothing to shortlist you with. It’s not penalising you. It simply has no material that maps to the question being asked, so it reaches for the sources that do, and those sources name your competitors.
This is the gap that costs the most and shows up the least. You can have a beautiful site, real expertise, and genuinely better outcomes, and still be missing from every shortlist, because shortlists are built from comparison and you’ve only ever published description.
Commercially, this is where it bites. Conductor’s 2026 data puts roughly 87% of AI referral traffic coming from ChatGPT, so this is concentrated, not scattered across a dozen engines you can safely ignore. And these people convert. Similarweb estimates ChatGPT referrals convert at 11.4% in ecommerce against 5.3% for organic, because the engine has already filtered for fit before the person ever lands on you. HubSpot’s State of AEO 2026 found 44% of marketers have made a business purchase based on a brand they discovered in an AI answer. Being absent from the shortlist is lost revenue you can’t see in any dashboard.
Each layer needs a different kind of content
The mistake is trying to fix all three with the same move, usually “publish more.” Volume doesn’t help if the content is the wrong shape for the layer you’re failing.
For informational, be the clear, structured explainer of your category. Answer the “what is” and “how does” questions properly, with definitions an engine can lift cleanly. This is the citeability layer, and it’s what most AEO advice online is actually about. Useful, but it gets you mentioned in the category rather than onto a shortlist.
Consideration is the one to fix first, because it’s where you’re most likely bleeding and least likely to know it. You need content that does the comparing for the engine. Who is this for, specifically. What situation does it suit. How does it compare to the obvious alternative, including the alternatives that aren’t competitors at all. So not “we offer ops consulting” but something closer to: for a founder-led agency past two million that’s drowning in undesigned complexity, here’s how this approach differs from hiring a COO or buying another tool. You’re handing the engine the exact language it needs to put you on the list.
Conversion is mostly already handled if your site and profiles are clear and consistent. The fix here is consistency of your entity across the web, so that when someone checks your name, every source agrees on who you are and what you’re best at. Contradictory or thin information is the usual failure mode, and it’s usually quick to sort out.
Notice the order. Most people pour effort into informational content and assume consideration takes care of itself. It’s backwards. Conversion is usually fine, consideration is usually the hole, informational is the long game. Fix the hole first.
How do I test the three layers of AI visibility myself?
You don’t need a tool to find out where you stand. Ten minutes and a willingness to read an honest answer will do it.
Open ChatGPT, because that’s where the traffic is, and run three rounds of prompts as if you were your own ideal client. Use their language, not your marketing language.
- Informational. “What is [your category]?” and “What should I look for in a [your type of provider]?” Are you, or your kind of approach, represented at all?
- Consideration. “Best [your category] for [your specific client’s situation].” “Alternatives to [the one competitor everyone’s heard of].” “[Competitor] versus other options for [situation].” Watch closely. Are you on the list? Who is, instead of you?
- Conversion. “Is [your business] any good?” “Tell me about [your business].” This one will probably look great. Don’t let it reassure you. It’s the easiest test and the least meaningful.
The tell is simple. If you only appear when you type your own name, you’re invisible at the exact moment people build their shortlist. You’re passing the test that doesn’t count.
That layered read is what the free Does AI Recommend You? check and the Citeability Snapshot both measure. The free check gives you a rough position across the layers in a couple of minutes. The Snapshot runs your real brand across the engines on real client prompts at each layer, shows you who’s being recommended instead of you, and ranks at least eight fixes to start on. Same logic as the ten-minute test above, done properly, with receipts.
The layer you’re best at is the one that matters least
Most service businesses are graded on the wrong test. They check whether AI can describe them, breathe a sigh of relief, and miss that the client’s decision happened one layer earlier, in a room they weren’t in.
Found, recommended, chosen. Three different jobs. You can be perfectly describable and still never make a shortlist, and a potential client can run their whole decision through an engine without your name surfacing once. The work isn’t to be known by AI. It’s to be on the list at the moment the list gets made.
Run the ten-minute test today. If you only show up when you type your own name, you’ve found your gap. If you want it measured properly, with the engines run on your real client prompts and the first fixes named, that’s what the Citeability Snapshot does. $197, done for you, no retainer attached.