Why AI forces a bottom-up acquisition strategy

The industry has been building top-down for 30 years. Start with awareness, get in front of as many people as possible, and work them down through the acquisition funnel.

The logic made sense in the broadcast era, and it wasn’t entirely wrong in the search era.

In AI-driven environments, it’s simply wrong.

Search engines, assistive engines, and agents build their ability to recommend your brand from the bottom up. They need to understand who you are before they can evaluate whether you’re credible. They need to evaluate your credibility before they recommend you to anyone.

If you build from the top down, you’re wasting budget on awareness while the engines and agents have no foundation to attach it to.

Agential systems make the stakes absolute. An agent acting on behalf of a user evaluates your brand, your offers, and your credibility, then commits.

If the machine doesn’t understand who you are, what you offer, and whom you serve, the agent can’t act in your favor. If it understands you but doesn’t find you the most credible option, it selects your competitor.

This is the ultimate zero-sum moment in AI: the recommendation you never saw happening, to the prospect you never knew was considering.

The acquisition funnel runs simultaneously in opposite directions

The user experience of the acquisition funnel hasn’t changed. Someone hears about you, considers you, and decides whether to commit. That journey runs wide to narrow, top to bottom: awareness first, evaluation second, and decision at the bottom.

This is the familiar funnel. Elias St. Elmo Lewis formalized it in 1898. Every marketing model since has been built around it, and for 128 years, nothing fundamental has changed. The channels evolved, but the direction was always the same: reach first, relationship second, commitment third. 

In 2002, my friend Philippe Lanceleur described the web perfectly for search: building a website and hoping people find it is like opening a shop in the middle of a field. Nobody passes by accident. You go where your audience hangs out, engage with them, and invite them to cross the field and visit your shop. Awareness was still the prerequisite, and your marketing had no chance of working without it.

The shift to entities changed the prerequisite. When Google introduced the Knowledge Graph in 2012, the machine began forming opinions about brands independently of what users were searching. The machine was drawing its own map and building roads for you. 

Those machine-built roads are built from the shop outwards by the machines, which means brand understanding and reputation, not awareness, become the prerequisite. All my work since 2012 has been focused on brand understanding and reputation for exactly this reason.

AI makes the acquisition funnel flip more powerful still. Assistive engines and agents now actively direct users toward destinations they’ve assessed as credible. Lanceleur’s shop in the field is no longer a handicap if the machines know it’s there and believe it’s the best destination for their users: they provide the roads.

This is the first genuine structural break in how brands must think about marketing since 1898. The display funnel is unchanged: the user still travels from awareness to decision. What makes you a candidate at the top of that funnel in AI engines and agents is built by training the machine to bring users to you.

How top-down and bottom-up coexist

The big takeaway is that the build funnel runs in the opposite direction. 

• The machine starts at the bottom. Does it know who you are? 
• It works up through credibility. Does it trust what you do? 
• Only then does it reach advocacy. Will it recommend you proactively? 

The moment of commitment by the user stays the same: know-like-trust the brand, but the only way for the user to arrive at that moment in AI assistive engines is that the machine knows, likes, and trusts your brand.

The coexistence of the bi-directional funnel is real. You can build top-down in channels you control: paid media, broadcast, and direct outreach. You can still buy awareness and pull people to decision. In the engines themselves, the user still has the top-down experience. 

The difference is that within the engines for organic, you have to build from the bottom of the funnel (BOFU) up because that’s how the machines build the roads to your brand.

Every algorithm, assistive engine, and agent operates on entity and brand signals, not on how loudly you push. Reach on social media has always been influenced by brand recognition, engagement, and topic, and here too, brand understanding and trust are gaining increasing weight.

With AI, roads to your shop in the field are increasingly machine-built, and machine-built roads are built from brand understanding outwards to awareness.

The original 1898 funnel still describes what users experience. In AI assistive engines and agents, it no longer describes the strategy that gets you in front of them: for that, you need to flip the funnel.

In short, you can’t build your funnel in AI engines and agents top-down in a world where those machines are the mediators between you and your audience. The machine won’t recommend brands it doesn’t understand, and it will only advocate for brands it trusts. This is a mechanical fact.

AI infrastructure works like this, so you also must. 

• Understandability creates the entity node.
• Credibility gives it preferential consideration.
• Deliverability gives it visibility.

Foundation. Proof. Reach. Put like that, it really does seem obvious, unavoidable, and comfortable.

How the funnel becomes a guided sequence in AI

The user journey on Google used to be a series of single-composed SERPs that users navigated themselves. Search engines composed those pages cleverly (Google and Bing have run a whole page algorithm since universal search launched in 2007, Darwinistically pulling elements from across verticals and scoring the composition as the “product”), but the navigation across the funnel was the user’s job.

As an SEO, you optimized for a position in the composition, and the user carried themselves from awareness to consideration to decision by browsing, comparing, and choosing.

Over the last few years, the algorithmic trinity has fundamentally changed that dynamic. The LLM reasons about what the user is asking, decides whether to answer directly, ground, search, or fact-check via the knowledge graph, and runs fan-out queries to retrieve across multiple angles of the question.

Those fan-out queries (which I’ve also called cascading queries) help the assistive engine answer the question more completely and more accurately than a single query would. But the breadth of what it gathers also lets it do one more thing — and this is the mechanic that actually matters in the funnel that leads to the perfect click: it can anticipate what the user is likely to do next, and set the current answer up to flow toward it.

The explicit representation of the LLM’s prediction of “next step” is the follow-up questions you see in the results. But there’s an additional implicit side to this architecture you might have missed: the way it composes the current answer shapes what the user is likely to do next. The AI is, to a very large extent, defining the acquisition journey. It seems to me the user is less in control than they feel.

That means your job appears to be to fight for a slot in a sequence the machine has already built.

That’s fair. But I’d argue that the brand’s job is also to train the machine’s expectations about what a logical next step looks like, so that when the LLM composes, your content is the natural thing it reaches for. 

You supply the ideas, you structure the follow-ups, you publish the logical bridges (“if you’re thinking about X, the next thing to consider is Y, and here’s the evidence”) in enough places, and with enough corroboration, that the machine treats those bridges as settled, not speculative. The machine then guides users toward you because your content is what its prediction landed on, because your framing is what made that prediction logical in the first place.

Now, is the AI thinking one step ahead? Or playing chess and planning several moves in advance? It depends. How far ahead the machine can usefully look depends on the territory. 

On well-traveled ground, the paths are well-worn, and the branches are narrow, so the LLM can stage two, three, or more moves ahead. Think of this as established neurological synapses: your influence on the paths is limited here. 

In unusual territory, the branches collapse the prediction horizon back to one, perhaps two steps. That’s an opportunity for a brand to create the synapses with your brand firmly anchored. Here’s yet another good reason to niche down, solve very specific problems, and have a very clear funnel pathway.

When defining the content I work on and terms I track, I use the concept of funnel pathway for exactly that reason — a top-of-funnel (TOFU) query that naturally leads to my brand at BOFU with a series of steps that are logical and relatively predictable.

So, track a set of terms that have a natural pathway to your brand at the zero-sum moment at the bottom of the funnel. Some start at TOFU and move through MOFU to BOFU. Others begin at MOFU with a clear path to BOFU, and some start (and end) at BOFU.

I’ll probably get pushback here. The number of possible paths is effectively infinite because conversations with AI can go anywhere. True. But this is a better system than chasing search volume or tracking the terms the boss likes: it forces you to think, focus, and prioritize — and it works.

Strategically, you have to get a foot in the door as early as possible in the conversation, and ensure that you keep your foot there as the conversation evolves and the AI guides the user down the funnel.

The stronger your foot in the door, the more you shape the conversation the machine builds, the more that conversation thins the field of competitors the machine considers for the next step, and, by virtue of elimination, the more likely you are to get the perfect click at the zero-sum moment at the bottom of the funnel.

I’m advocating for educating the algorithms (remember, Google is a child?). The better you guide, the more the machine’s best-brand prediction converges on you step after step, because the path it’s following is the path you built into its brain. 

Get in high, and the compounding works in your favor. Get in late, and your competitors’ bridges become the machine’s bridges, and every subsequent step is a fight to re-enter a sequence where your competitor is Top of Algorithmic Mind.

Display is where your acquisition funnel lives in the AI engine pipeline

The AI engine pipeline runs 10 gates from discovered to won. 

• Everything up to annotation (Gate 5) is infrastructure: can the machine access, store, and classify your content? 
• From recruitment (Gate 6) onward, the engine compares you to every alternative. 
• The understandability, credibility, and deliverability (UCD) layer is where the user sees the machine evaluation at display (Gate 8). Understandability is the key to won (Gate 9).

The three dimensions of brand visibility at display

Display is the moment when the machine can make or break your brand by being the most visible in the market at every touchpoint when your ideal customer profile (ICP) is having a conversation with the engine or agent. 

It’s obvious that this is the key moment when you need the engine or agent to be absolutely convinced that you’re the best solution to the specific user’s problem at the exact moment they convert (see the 95/5 rule here).

Understandability (U) is the trusted partner/decision layer, without which nothing else will work long term. Does the machine know who you are, what you do, and who you do it for? 

U is BOFU, which is both the moment of decision and (logically) the deepest trust layer for both the AI user and the human user. When someone searches your brand name or asks an AI assistant directly about you, the machine draws on its understanding of your entity. 

If that understanding is weak, contradictory, or absent, the machine either hedges or stays silent. Typical failure modes show up in AI responses as “claims to be,” “appears to offer,” or “no idea who you are talking about.” The doubt tax — where prospects ready to buy get a hedge instead of a confirmation — is a U failure.

Credibility (C) is the recommender/consideration layer. Does the AI believe you’re genuinely better than your competitors at what you do? 

C is MOFU, the comparison and evaluation layer. When someone asks an AI who is the best in market, the machine draws on its confidence in your N-E-E-A-T-T credibility and will exclude you if you haven’t built a rock-solid argument to be cited. 

If AI confidence in you is weaker than its confidence in the credibility of your competitor, you lose the comparison. The ghost tax – absent from competitive evaluation and ignored in shortlists — is a C failure.

Deliverability (D) is the advocate/awareness layer. Does the AI surface your brand to people who aren’t searching for you, recommend you unprompted when they research the market, and treat you as the reference option in your category? 

D is TOFU, the reach layer. When someone asks an AI about a problem, you solve without knowing your brand exists, the machine draws on its confidence that you are the right answer to put in front of them. 

Advocacy only happens when the machine has first understood who you are (U), and judged you better than the alternatives (C). The invisibility tax — never mentioned to prospects researching the market — is a D failure.

The business case for UCD: The three taxes

My untrained salesforce framing is super clear for a non-technical audience. Google, ChatGPT, Perplexity, Claude, Copilot, Siri, and Alexa are seven employees working 24/7, and they’re either selling for your brand or for your competitors. AAO can be defined as training AI assistive engines and agents to sell for you at the top, middle, and bottom of the funnel.

Here’s the part most of the industry still hasn’t internalized: machines aren’t an alternative audience. They’re a mirror of how people process information, with the noise filtered out. 

Optimizing for machines is optimizing for humans with less guesswork. A brand SERP is Google’s opinion of the world’s opinion of you, and Google’s opinion is built from the same signals that form human opinion, only weighted more consistently, and corroborated across millions of data points. 

When you optimize to improve what Google believes about your brand, you’re not gaming an algorithm. You’re correcting and reinforcing what the world already believes about you, expressed with the precision humans rarely articulate. The algorithm is the clearest feedback loop marketing has ever had. 

Each tax is a specific failure mode of that untrained salesforce. 

• The doubt tax is what you pay when they can’t confirm who you are to a prospect ready to buy. 
• The ghost tax is what you pay when they can’t argue your case against competitors in a shortlist. 
• The invisibility tax is what you pay when they don’t mention you at all to the prospect researching the market. 

The fixes run in one order: U before C, C before D, because the taxes are mechanically ordered, and the remediation has to match.

Content was king in the keyword era, context took the throne around 2016, and confidence is king now. The AI engines don’t just store and retrieve. They stake their own credibility on the brands they recommend, and that staking runs on accumulated confidence at every layer. 

Build U to retire the doubt tax. Build C to retire the ghost tax. Build D to retire the invisibility tax. Every tax retired is a recommendation earned, and every recommendation earned is revenue the machine now generates on your behalf instead of your competitor’s. 

Strategy: Your brand SERP and AI résumé tell you where to begin

Brand SERP is what Google shows when someone searches your brand name. The AI résumé is the same object in conversational format. The agent dossier is the machine’s silent judgment during evaluation before any recommendation reaches a person. 

All three are dual-function objects. They’re the machine’s output to every audience that asks about you, and your diagnostic instrument for reading the machine’s current confidence. That dual function is why they’re both the product and the audit.

Read all three as the machine’s understanding of you, its assessment of your credibility, and its confidence in you as a solution provider. The diagnostic triage is short.

If the machine gets things wrong, hedges facts, or the results don’t reflect your brand narrative, that’s an understandability problem. The entity record is inconsistent, weak, or contradictory, and the work is on your entity home: clean structured data, consistent descriptions, clear schema, and entity resolution that points to a single authoritative source.

If the results are unconvincing, unflattering, or don’t do you full justice, that’s a credibility problem. Your N-E-E-A-T-T is weak, and the work is offsite: third-party mentions, review platforms, earned media, and co-citations from sources the machine trusts.

If the results don’t reflect your digital marketing strategy, that’s a deliverability issue. The work is in content, both on your channels and on third-party properties, the type of material the machine treats as proof rather than a claim.

In every case, the diagnosis comes before the tactics. U before C, C before D, and the sequence isn’t optional.

Acquisition is one act in a 15-stage play

The acquisition funnel feels dominant because it’s where conversion happens. The funnel sits on the display gate, where UCD determines whether the machine recommends you. 

Everything else, the work that lets display happen at all and the work that compounds afterward, runs across the nine gates before it and the five gates after it.

Those five gates after Won are where most of the money is made and most of the confidence is generated. Onboarded, performed, integrated, devoted, and codified — every client outcome feeds signals back into gate zero for the next prospect who has never heard of you. 

The flywheel is the mechanism. Get it right, and every satisfied client strengthens the machine’s confidence in your brand for the next one. Get it wrong, and every neutral outcome decays it.

That’s more than just an acquisition strategy; it’s a business strategy, with the machine as a constant participant at every stage.

The final articles in this series will show you what happens after won: how every satisfied client either trains the machine to recommend you more confidently next time, or quietly erodes the confidence you’ve already built. 

The funnel isn’t where the money is made, but it is the critical moment the flywheel feeds where the path to money is.

This is the 10th piece in my AI authority series. 

• The first, “Rand Fishkin proved AI recommendations are inconsistent – here’s why and how to fix it,” introduced cascading confidence. 
• The second, “AAO: Why assistive agent optimization is the next evolution of SEO,” named the discipline. 
• The third, “The AI engine pipeline: 10 gates that decide whether you win the recommendation,” mapped the full pipeline. 
• The fourth, “The five infrastructure gates behind crawl, render, and index,” walked through the infrastructure phase.
• The fifth, “5 competitive gates hidden inside ‘rank and display’,” covered the competitive phase.
• The sixth, “The entity home: The page that shapes how search, AI, and users see your brand,” mapped the raw material.
• The seventh, “The push layer returns: Why ‘publish and wait’ is half a strategy,” extended the entry model. 
• The eighth, “How AI decides what your content means and why it gets you wrong,” covered annotation — the last gate where you’re alone with the machine. 
• The ninth, “Why topical authority isn’t enough for AI search,” opened the competitive phase proper with topical ownership.
• Up next: Why evidence on its own isn’t enough, and how the framing gap explains which brands AI recommends and which it hedges on.