Why SEO agencies fail at AI search.
I want to be careful here, because “SEO agencies are obsolete” is a cheap thing to say and it is not what I think. Good SEO people are good at something genuinely hard, and most of what they know still applies.
But I keep seeing the same engagement fail the same way, and the failure is structural rather than a matter of competence. It is worth naming precisely, because if you are paying for this work you can check for it in an afternoon.
The structural difference
A search engine returns a ranked list. Being eleventh is worse than being first but it is not nothing — you exist, you are reachable, you accumulate. The entire discipline is built on that gradient: move up, capture more.
An assistant returns one answer naming perhaps three businesses. There is no eleventh. You are in the answer or you are not, and the person asking never learns you existed.
Every structural difference that follows comes from that one fact.
| Search | AI answer | |
|---|---|---|
| Result shape | Ranked list, ten-plus results | One synthesis, a few names |
| Partial credit | Yes — position 11 still gets traffic | No — named or invisible |
| Unit optimised | A page against a query | A business as an entity |
| What wins | Relevance plus authority to the page | Independent sources agreeing about you |
| Self-description | Counts — it is your page | Weakest available evidence |
| Feedback | Rankings, impressions, clicks | None by default; you must go and ask |
The five failures, in order of how often I see them
1. Optimising pages when the unit is the business. SEO is organised around pages: this page targets this query. AI visibility is organised around the entity. A model deciding whether to name you is not evaluating a page, it is evaluating whether it knows enough consistent things about your business to say something. An agency can lift twenty pages and leave the entity exactly as unresolvable as before.
2. Producing content instead of corroboration. The standard answer to any visibility problem is more content. But content on your own site is you talking about yourself, and that is the least persuasive evidence a model has. What moves the needle is independent sources agreeing about you. That work looks like outreach, listings, press and partnerships — not a content calendar, and it is not what most retainers are staffed for.
3. Reporting on metrics that no longer describe the outcome. Rankings, impressions, clicks. All real, all still worth watching, none of which tells you whether an assistant names you. A monthly report can be entirely green while you are absent from every AI answer in your category, and nothing in the report would reveal it.
4. Treating technical hygiene as finished. The technical items that matter for AI retrieval overlap with SEO but are not identical. Content that renders only after JavaScript is the clearest case: it is fine for Google, which executes JavaScript, and invisible to AI crawlers, which largely do not. A site can pass every technical SEO audit and still be substantially unreadable to the systems you are trying to reach.
5. Never measuring the actual outcome. This is the one that lets the other four persist. If nobody asks an assistant what it says about the business, nobody finds out the programme is not working on that axis. The measurement is not hard — open an assistant, ask the questions a customer would, write down what comes back and which sources it used. The absence of that habit is the real tell.
Notice that none of these is about talent. They are about what the engagement was designed to produce. An agency hired to improve rankings, measured on rankings, will improve rankings. That is the system working.
What good looks like instead
The parts of SEO that transfer, transfer completely. Technical crawlability, clear information architecture, genuinely useful content, earned links — all still load-bearing. If your foundations are bad, fix them first; AI visibility work on a broken site is decoration.
What gets added:
Entity consistency. The same facts about your business everywhere it appears. Tedious, unglamorous, and the highest-yield item for most businesses.
Third-party corroboration, deliberately built. Identified by measurement — find which sources the assistants actually use for your category, then work on being present and accurate in those — rather than by assumption.
Content that states its evidence. Not more content. Content that makes a claim, shows the support, and names the source. That is what gets quoted, and it is a different writing brief from what ranks.
Measurement against assistants directly. Monthly, same questions, recorded. The number matters less than the direction and the source list.
How to check your own engagement
Three questions. They take an afternoon and they are not hostile — a good agency will have answers.
“When you last asked an assistant about us, what did it say, and which sources did it use?” If nobody has asked, nobody is working on this.
“Which third-party sources are we trying to appear in, and why those?” A good answer names specific sites and explains how they were identified. A weak one talks about authority in general.
“What are our facts, and where do they disagree?” If nobody has inventoried name, address, phone, category and hours across every listing, the entity work has not started.
If the answers are thin, that is not necessarily a reason to leave. It is a reason to be explicit that you now want something the engagement was never scoped to deliver — and to add the measurement, because without it neither of you will know whether anything changed.
Start with the measurement. The free audit asks a live assistant real customer questions and reports which sources it used to answer.
Run the free auditCommon questions
Why doesn't my SEO agency get me cited by AI assistants?
Usually because the engagement optimises something different. SEO competes for position in a ranked list, where being eleventh still produces traffic. An AI assistant returns one answer naming a few businesses, so there is no partial credit. The work that produces inclusion — consistent facts about the business across every source, and corroboration from sites you do not control — is not what a retainer scoped around pages and rankings is staffed to do.
Is SEO still worth paying for?
Yes. Technical crawlability, information architecture, useful content and earned links remain load-bearing for both search and AI retrieval, and a site with weak foundations will not perform in either. The issue is not that SEO stopped working; it is that it is necessary and no longer sufficient, and most engagements have not added the entity and corroboration work that AI visibility requires.
Can a green SEO report coexist with being invisible to AI?
Yes, and it is common. Rankings, impressions and clicks can all improve while an assistant never names the business, because none of those metrics measures inclusion in a synthesised answer. The only way to detect it is to ask assistants directly, using the questions a customer would ask, and record both the answer and the sources cited.
What should I ask my agency about AI visibility?
Three things. What did an assistant say about us when you last asked, and which sources did it use? Which third-party sources are we working to appear in, and how were they identified? And what are our business facts, and where do they currently disagree across listings? Thin answers to all three indicate the work has not begun, regardless of what the monthly report shows.
Does publishing more content improve AI visibility?
Less than expected. Content on your own site is self-description, which is the weakest evidence available about you. What moves AI visibility is independent sources agreeing about your business, plus content that states a claim and shows its supporting evidence rather than asserting it. Volume alone tends to produce more pages that nobody cites.
About the author. Siamak Kalhor is an AI marketing and LLMO consultant in Los Angeles with more than forty years of experience in marketing strategy. He advises professionals and e-commerce leaders on becoming the answer AI assistants give. Get in touch, or connect on LinkedIn.