You're reading an RFP response. It arrived eleven minutes before the deadline, which is the first red flag, but you've learned to ignore that one because all agencies run on deadline fumes the way the rest of us run on oxygen. The cover page has a new logo treatment - they've added a gradient - and the agency's name now includes the word "Intelligence," which it did not include the last time you saw a proposal from them, which was eight months ago, when they were still just a name and an ampersand and another name. You scroll past the capabilities section, past the case studies you've already seen, past the team bios that still include that one guy's marathon time, and you arrive at the new offering. It's on page fourteen. "AI Search Optimization." It has its own section. It has its own pricing tier. It costs 40% more than their regular SEO package.
You lean forward in your chair. You have been waiting for someone to show you something real about AI search, because you know it matters, because your CEO read a Wired article three weeks ago and now brings it up in every leadership meeting with the energy of someone who has recently discovered cold plunge therapy. You need an answer. You need a strategy. You need someone who knows what they're doing.
You look at the deliverables.
They are identical to the regular SEO package. Not similar. Not "enhanced." Identical. Monthly content production. Technical audits. Keyword research. Backlink analysis. Quarterly reporting. The same twelve line items you've been reading in SEO proposals since 2014, except now each one has the word "AI" inserted before it like a prefix that was stamped on at the print shop. "AI Content Strategy." "AI Keyword Research." "AI Technical Optimization." The content team has been renamed the "AI Content Intelligence Team." The reporting dashboard has been renamed the "AI Visibility Dashboard." The quarterly strategy call has been renamed the "AI Strategy Alignment Session."
The work is the same. The people are the same. The process is the same. But now there's a surcharge.
It's like your plumber adding "smart home plumbing" to their business card. The pipes are the same. The wrench is the same. The guy is the same. He still shows up between 9 and 4, which means he shows up at 3:47. But now when he snakes your drain, it's "smart drain optimization," and when he replaces your faucet, it's an "intelligent water delivery endpoint," and the invoice is 40% higher because intelligence, as it turns out, costs extra, even when it's just a sticker on the same toolbox.
This is happening everywhere. I have reviewed nine agency proposals in the last four months that include an AI search tier, and I want to tell you how many of them contained a single deliverable that was genuinely specific to how large language models retrieve and surface information. The number is one. One out of nine. And that one was from a boutique shop run by a former ML engineer who pivoted to SEO, which is not the typical agency origin story (the typical agency origin story involves a WordPress blog, a burst of link building in 2011, and the discovery that you could charge people monthly for something that used to be a one-time project).
The Relabeling Industrial Complex
I have been in SEO for more than twenty years, and in that time I have watched the industry rename itself approximately once every thirty months. We were search engine marketers, then we were inbound marketers, then we were growth hackers (briefly, mercifully), then we were content strategists, then we were "full-stack" something, and now we are AI search optimization specialists. Each rename coincides with a repricing event. The rename is the strategy. The deliverables trail behind it, unchanged, like a dog tied to a truck that's been repainted.
I'm not even angry about it, really. I understand the economics. Agencies operate on margins that would make a grocery store look like a hedge fund, and when a new wave of client demand appears - when every CMO starts asking "what are we doing about AI search" the same week - you have approximately ninety days to have an offering or you lose the RFP cycle entirely. Ninety days is not enough time to build genuine capability. It is enough time to build a slide deck. So that's what gets built.
The slide deck gets a title. The title gets a price. The price gets approved because the CMO needs to tell the CEO they have an AI strategy, and the CEO needs to tell the board the same thing, and the board needs to tell the investors, and so a chain of reassurance extends from the agency account manager all the way up to a limited partner who read the same Wired article your CEO did, and everyone in the chain is satisfied because everyone in the chain has the same need, which is not "results" but "a story about results that sounds credible in a meeting."
This is not a conspiracy. It's an ecosystem. Every organism in it is behaving rationally given its incentives. The agency needs revenue. The CMO needs air cover. The CEO needs a narrative. The investor needs a thesis. AI search provides all four with a single phrase. The fact that nobody in the chain actually understands how LLMs retrieve information is, from an ecosystem perspective, not relevant. The ecosystem does not select for understanding. It selects for plausibility.
What They Don't Know (And Can't Sell You)
Here is the core problem, and I'm going to be specific about it because specificity is the thing that's missing from every AI search pitch I've evaluated.
When a large language model answers a user's question, it is doing something fundamentally different from what Google does when it returns ten blue links. Google matches query terms to document terms, weighted by a thousand signals including link authority, content quality, user engagement, and (increasingly) semantic understanding. The process is retrieval-first: find the documents, then rank them. The user sees the documents. The user clicks. The click is the value.
An LLM does not do this. An LLM generates a response by predicting the next token in a sequence, drawing on patterns absorbed during training and - in the case of retrieval-augmented generation - on documents pulled from a search index in real time. The output is not a ranked list of documents. It is a synthesized answer. The user may never see your document. Your document may have informed the answer without being cited. Your document may have been cited but in a way that answers the question so completely that nobody clicks through. The value chain is structurally different.
Most SEO agencies do not understand this. I don't mean they haven't read about it. I mean they don't understand it at the level where understanding translates into different work. They're still thinking in keywords. They're still thinking in rankings. They're still thinking in click-through rates and landing pages and conversion funnels that begin with a search result and end with a form fill. That entire mental model - the model on which their processes, their tools, their hiring decisions, and their pricing are built - assumes a user who sees a list of results, chooses one, and clicks. And for traditional search, that model still works. But for AI search, that model is like using a roadmap to navigate the ocean.
Let me give you a concrete example. An agency tells you they'll optimize your content for AI search. You ask them how. They say they'll do "comprehensive content" that "thoroughly covers the topic" so that LLMs will "prefer your content as a source." This sounds reasonable. It is also exactly what they would have told you about regular SEO. The deliverable is a 2,500-word blog post with an H1, five H2s, a meta description, and some internal links. It is content marketing. It has always been content marketing. The AI label is decorative.
What they can't tell you - because they don't know - is how an LLM decides which source to cite when generating a response. They can't tell you about entity recognition and why being a well-defined entity in the knowledge graph matters more for AI retrieval than any individual piece of content. They can't tell you about the specific structured data patterns that make your information machine-readable in a way that influences training data incorporation. They can't tell you about citation-worthiness - why some sources get cited and others get synthesized without attribution - because the research on this is six months old and published in ML papers, not marketing blogs, and nobody at the agency reads ML papers because nobody at the agency was hired to read ML papers.
They don't know what they don't know. And the gap between what's being sold and what's being delivered is not a small gap. It is the kind of gap that wars are fought in.
The Keyword Hangover
The deepest problem is conceptual, and it goes back to the beginning of the industry. SEO was built on keywords. The entire practice, from its origins in the late 1990s through its maturation in the 2010s, was organized around the idea that people type queries into a box, and you optimize pages to match those queries. The tools are keyword tools. The strategies are keyword strategies. The reports track keyword rankings. The pitches lead with keyword opportunities. The entire cognitive framework of the profession is a keyword framework.
AI search does not work on keywords. I need to say this clearly because I keep seeing agencies apply keyword thinking to LLM systems and it is like watching someone try to unlock a door with a fishing rod - technically you're interacting with the door, but the mechanism of interaction is entirely wrong.
When someone asks ChatGPT "what's the best SEO consultant for mid-market B2B companies," the model is not matching keywords. It is synthesizing a response based on its understanding of the entities "SEO consultant," "mid-market," and "B2B," combined with whatever evidence exists in its training data or retrieval sources about specific consultants who match those criteria. The model is reasoning (or performing something that looks enough like reasoning that the distinction is academic). It is not retrieving a ranked list of keyword-optimized pages.
This means the entire SEO toolkit - keyword research, keyword mapping, keyword density analysis, keyword gap analysis, keyword difficulty scores - is not wrong for AI search, but it's not right either. It's operating at the wrong level of abstraction. It's like using a street map when you need a topographic map. The territory is the same, but the features that matter are different. For AI search, what matters is: Are you an entity? Are you a well-defined, well-connected entity that the model recognizes as authoritative on specific topics? Do you have structured data that makes your attributes machine-readable? Have you produced content that is citation-worthy - content that contains original data, original analysis, original frameworks, things that can't be synthesized from other sources because they didn't exist until you created them?
These are not keyword questions. They are entity questions, authority questions, originality questions. And most agencies are not equipped to answer them because their entire operation - the tools they pay for, the people they hire, the processes they run, the templates they use - is a keyword operation.
The Content Marketing Shell Game
I need to talk about what agencies mean when they say "AI content strategy," because I have now seen this phrase in enough proposals that I feel qualified to translate it.
"AI content strategy" means content marketing. It means blog posts, pillar pages, topic clusters, and editorial calendars. It means the same content strategy they were selling before, targeted at the same keywords (yes, keywords - they haven't actually changed the targeting methodology, just the label), distributed through the same channels, measured by the same metrics. The "AI" part of "AI content strategy" refers to the fact that the content might show up in an AI-generated response, the same way any content might show up in an AI-generated response, the same way any content might show up in a Google snippet, which is not a strategy. It is a hope. Hopes are fine. Hopes are not worth a 40% premium.
Here is what an actual AI search strategy would involve, if one could be built with today's tools and data (and I want to be honest: a fully realized AI search strategy is still somewhat speculative, because the measurement tools genuinely do not exist yet, which is itself a reason to be skeptical of anyone selling one with certainty).
First, entity optimization. Not in the SEO-buzzword sense of "we'll update your Knowledge Panel." In the deep sense of ensuring that your brand, your people, your products, and your expertise are represented as clearly defined entities across the web - in Wikipedia, in industry databases, in professional profiles, in structured data on your own site, in the corpus of content that LLMs are trained on. This is not a content strategy problem. It is a digital presence problem, closer to PR than to content marketing.
Second, citation-worthiness. Producing content that contains things worth citing - original research, proprietary data, novel frameworks, expert analysis that cannot be found elsewhere. Not "comprehensive guides" that summarize what already exists, but source material that other people (and other machines) cite because it contains something new. This requires subject matter expertise, which most agencies do not have, because most agencies are staffed by generalists who can write competently about anything but are experts in nothing (which is a structural feature of the agency model, not a personal failing - you can't hire deep experts at agency rates).
Third, structured data at a level most agencies do not implement. Not just schema markup on your blog posts (which they're already doing, sort of, when they remember). I mean comprehensive structured data that makes your entire information architecture machine-readable - organization schema, person schema, product schema, FAQ schema, how-to schema, dataset schema. The kind of implementation that requires a developer, not a content writer, and that most agencies outsource to a freelancer who implements it once and never touches it again.
Fourth, and this is the one nobody wants to hear: accepting that for some queries, in some categories, the goal is not to get the click but to be the source. Being the entity the model references, the brand the model names, the authority the model defers to - even if the user never visits your site. This requires a fundamental shift in how you measure success, and agencies cannot make that shift because their entire business model is built on reporting metrics that require clicks. An agency that says "we improved your AI visibility but traffic went down" will lose the account. So they don't say it. They track the same metrics. They report the same numbers. The AI label is decoration over a fundamentally unchanged practice.
The People Who Actually Get It
I know this sounds like I'm saying nobody can do AI search well. That's not what I'm saying. Some people do it very well. They are just not the people selling "AI Search Optimization" as a service tier.
The people who actually understand how to perform well in AI search are, overwhelmingly, the same people who were already good at SEO. Not because AI search is the same as regular SEO (I wrote a whole piece about how the foundations overlap but the nuances differ). Because the skillset that makes you good at SEO - deep technical understanding, genuine content expertise, the ability to think about information architecture at a structural level, the patience to build authority over years instead of sprinting for quick wins - is the same skillset that makes you good at AI search.
The best "AI search optimization" work I've seen in the past year was done by people who never called it that. A technical SEO consultant who implemented comprehensive structured data across a client's entire site, not because of AI search, but because structured data has been important since 2015 and she actually does it properly. A content strategist who focused on original research and proprietary data, not because of AI citation, but because original content has always outperformed derivative content. A brand strategist who built her client's entity presence across the web, not because of LLM training data, but because brand authority has been a ranking signal since Google started.
These people didn't rename their services. They didn't create new pricing tiers. They didn't add "AI" to their LinkedIn headlines. They just kept doing the work they were already doing, and that work happened to be exactly what AI search systems reward, because AI search systems are built on the same foundation as traditional search systems, and the people who built strong foundations don't need to rebuild them every time someone adds a new floor.
The gap is not between "AI SEO" and "regular SEO." The gap is between good SEO and bad SEO. Good SEO has always meant building genuine authority, producing content worth referencing, and ensuring your technical foundation makes your information accessible to machines. That's it. The machines changed. The foundation didn't.
What You Should Do Instead of Buying the Service Tier
If you're evaluating agencies or consultants for AI search readiness (and I realize the irony of me, a consultant, telling you how to evaluate consultants - I am aware that the wolf is writing the guide to wolves), here is what to ask.
Ask them to explain how retrieval-augmented generation works. Not in marketing terms. In technical terms. If they can't explain how an LLM decides which sources to pull during RAG, they don't understand the system they're claiming to optimize for. This is the equivalent of asking an SEO person to explain how Googlebot crawls a site. It's foundational. If they can't do it, walk.
Ask them what structured data they implement beyond basic article schema. If the answer is "we add FAQ schema to blog posts," that's fine, but it's not AI search optimization. It's regular technical SEO that most agencies have been doing (or claiming to do) since 2019.
Ask them how they measure AI search performance. This is the trick question, because the honest answer is "we can't, not reliably, not yet." If they show you a dashboard with confident numbers, those numbers are either fabricated or pulled from tools that are sampling from a stochastic system in ways that produce the appearance of data without the substance of it. The honest answer is uncertainty. Uncertainty is not what agencies sell. But it's what honest practitioners feel, because the measurement infrastructure for AI search genuinely does not exist yet, and pretending it does is either ignorance or dishonesty, and you don't want to pay for either.
Ask them what they would do differently for AI search that they wouldn't do for regular SEO. If the answer is nothing, they're honest but also proving that their "AI search tier" is a rebadged SEO package. If the answer is a specific, technical list of activities - entity optimization, citation analysis, training data auditing, structured data expansion, source authority building - they might actually know what they're talking about. But verify. Ask for examples. Ask for results. Ask them to show you a before-and-after where their work specifically improved AI citation, and then ask them how they isolated that variable, because it is nearly impossible to isolate that variable with today's tools, and anyone who claims otherwise is selling you something.
The Honest Version
Here is what I tell clients when they ask me about AI search, and I want to say it plainly because I have been sarcastic for three thousand words and the thing deserves plain language.
AI search matters. It is not a fad. It is not going away. The way people find information is genuinely changing, and if you ignore that change, you will lose ground to competitors who don't. This is real.
But the response to a real change is not to buy a fake service. The response is to do real work. Build your entity. Produce original research. Implement comprehensive structured data. Create content that is worth citing - not content that covers topics, but content that advances them. Be a source, not a summarizer. And do all of this as part of your regular SEO practice, not as a separate initiative with a separate budget and a separate team, because separation is how you end up paying twice for the same foundation.
The agencies will catch up. Some of them are catching up right now, building genuine capability behind the marketing language. Give it eighteen months and the good ones will be doing real work in this space, the same way the good ones eventually did real mobile optimization after years of selling "mobile-friendly" as a checkbox exercise. The market corrects. It just corrects slowly, and in the meantime, a lot of money gets spent on slide decks.
Your plumber doesn't need a new title. He needs to keep the water flowing and the pipes clean and the joints sealed. If he does that well, the smart home will work fine, because the smart home sits on the same pipes the dumb home did, and the water doesn't know the difference.
Neither does the machine.