AI SEO Agency: What the Term Covers, and How to Judge One

The two jobs the term covers

The first job is AI as the labor. The agency still does search engine optimization the way it has been done for twenty years — keyword research, content, site structure, links — but uses language models to do it faster and cheaper. The output is aimed at Google's traditional ranked list of blue links. Nothing about the goal has changed; only the cost of producing the work has.

The second job is AI as the audience. Here the aim is to be quoted when somebody asks ChatGPT, Perplexity, Claude or Google's AI Mode a question, rather than to rank in a list they may never scroll. This has picked up the name generative engine optimization, and it is a different discipline because the thing you are optimizing for does not publish rankings, does not report impressions, and does not give you a dashboard.

Most agencies advertising as AI SEO are doing the first and describing it in the language of the second. That is the single most useful thing to know before you take a sales call, because the first is close to a commodity — the tools are available to you at consumer prices — while the second involves work that is harder to fake and harder to buy.

Neither is disreputable. An agency that says plainly "we use AI to produce more content for less" is being honest about a real service. The problem is the agency that implies the second while delivering the first, and prices accordingly.

• AI as the labor — models produce the content, audits and briefs. Target: Google's ranked results. Genuinely faster, close to commodity.
• AI as the audience — the goal is being cited inside AI answers. Target: assistants and AI search. Newer, harder to measure, harder to fake.
• The tell: ask which one they mean, and what they would show you at ninety days to prove it worked.

What actually changed in search

The reason this category exists is that answers increasingly arrive without a click. The Pew Research Center tracked the browsing of 900 US adults across 68,879 Google searches in March 2025 and found that when an AI summary appeared, people clicked a traditional search result on 8% of visits, against 15% of visits where no summary appeared — roughly half as often. They clicked a link inside the summary itself on just 1% of visits, and ended their browsing session entirely on 26% of pages with a summary, against 16% without.

Those are Pew's figures for one month of one year, and worth holding as a direction rather than a constant. Every later measurement of this has produced a different number depending on the query mix studied, which tells you the effect is real and its size is unsettled.

The scale on the other side is not in dispute. Google reported at its own I/O event in May 2026 that AI Mode had passed one billion monthly users, with queries more than doubling every quarter since launch. That is Google's own number about Google's own product, so treat it as a company claim rather than an audited figure — but even discounted heavily, it describes a surface that did not exist three years ago and now sits in front of search for a very large number of people.

Put those together and the practical change is this: ranking and being seen have come apart. You can hold position three and lose traffic, because the answer was assembled above you. An agency worth paying has a clear position on that, and it is not "we will future-proof your rankings."

This is the ground we cover in an AI consultation — a diagnostic session that ends with a costed plan rather than a proposal.

When AI is doing the SEO work

Used carefully, models make parts of this job genuinely faster. Clustering thousands of queries into topics, drafting outlines from what already ranks, spotting internal linking gaps, turning a crawl into a prioritized list, writing schema markup, translating a page set — these are pattern tasks with checkable output, and that is exactly where this technology is strong.

The trap is content at volume. Publishing a hundred generated articles a month is the easiest thing an agency can do, the easiest thing to put on an invoice, and the least likely to work. Google's guidance has been consistent that the issue is not whether a machine was involved but whether the page is useful and original; a large volume of adequate, interchangeable pages is the definition of neither. The failure is rarely a penalty. It is more often that nothing happens at all, slowly, for six months.

There is also a cost the volume model hides. Pages that exist need maintaining, redirecting, and eventually pruning. An agency that leaves you with four hundred thin articles has not given you an asset; it has given you a maintenance liability that the next agency will quote to clean up.

So the question to ask is not whether they use AI. Everyone does. It is which parts of the work it touches, what a person checks before anything publishes, and whether they can show you a page they would defend on its own merits without mentioning how it was made.

• Reasonable: query clustering, crawl triage, schema, outlines from the current results, translation, internal link discovery.
• Reasonable with a person on it: drafts that a subject expert rewrites and fact-checks before publication.
• Rarely works: publishing at volume, spinning near-duplicate location pages, generated pages nobody reads before they go live.
• Ask what happens to the pages that do not perform. An agency with no pruning plan is building you a liability.

When the goal is being cited by AI

The newer work is making your business legible to systems that answer rather than list. Some of it is unglamorous and overlaps with ordinary technical SEO: pages that state what you do in plain sentences rather than in marketing abstractions, structured data that says who you are, content organized so a specific claim can be lifted out of it without the surrounding paragraph.

Some of it is different in kind. Assistants assemble answers from what they can find about you across the web, not only from your own site — so what third parties say, whether your details agree across directories and profiles, and whether anyone independent has written about you all feed the answer. That is closer to public relations and reputation work than to traditional on-page optimization, which is why agencies staffed only with content writers struggle with it.

The honest difficulty is measurement. There is no impressions report for an assistant. The workable substitutes are direct and manual: ask the major assistants the questions your buyers ask, record what they say about you and who they cite, repeat it on a schedule, and watch whether the answers improve. That is a real method, and it is unglamorous enough that agencies selling certainty tend to skip it.

Treat any promise of guaranteed placement inside an AI answer as disqualifying. Nobody controls what these systems say. The people doing this work properly talk in terms of making citation more likely and checking whether it happened, not in terms of positions won.

What is not worth paying for

A few things get sold heavily in this category and are worth refusing outright.

The most common is a retainer whose deliverable is a content quota. Thirty articles a month is an activity, not an outcome, and it is the easiest thing to produce and the hardest to connect to anything you care about. If the contract measures output rather than result, the incentive runs against you from the first invoice.

The second is any tool or package promising to make you "AI-proof" or guaranteeing inclusion in AI Overviews. There is no such mechanism. The third is a dashboard that reports an invented visibility score with no stated method — if you cannot find out how the number is produced, it is not a measurement, it is a reassurance.

And a structural one: be wary of an agency that wants to move your site onto its own platform or host it under its own accounts. That has little to do with AI and everything to do with how hard it will be to leave. Whatever the discipline, the work should live in property you own.

How to tell a real one from a repackaged one

The useful questions here are about measurement, not method — because the method is increasingly the same everywhere and the measurement is where the honesty shows.

Ask what they will measure, and what the number was before they started. An agency that cannot state today's baseline cannot demonstrate a change later, and the absence of a baseline at the start is the most reliable early sign of a report full of charts and no conclusion at the end.

Ask what a bad quarter looks like and what they would do about it. Anyone who has run this work for real has had one and can describe it plainly. Ask which parts of their process a person does, and what that person changes. Ask to see the questions they would monitor in the assistants on your behalf, which is a request an agency doing the second job can answer in a sentence and an agency doing the first cannot answer at all.

Then ask the awkward one: what would make you tell us not to do this? A vendor with no answer is selling, not advising. This is the same test that applies to any agency, and it is worth applying here precisely because the subject is new enough that confident language is doing a lot of work.

• What is the baseline today, in numbers, before any work starts?
• Which assistant questions will you monitor, and how often will you re-run them?
• What does a person review before anything publishes?
• What does a bad quarter look like, and what happens then?
• Who owns the site, the content and the accounts if we stop working together?

When you do not need an agency at all

A good share of businesses asking this question do not have a search problem. They have a positioning problem that shows up in search, and no amount of optimization fixes a page that never says clearly what the company does and who it is for.

The cheapest test is to read your own homepage as a stranger, then ask a couple of assistants what your company does and see what comes back. If the answer is vague or wrong, the fix usually starts with writing plainly about your business rather than with hiring anyone. It is ordinary work, it costs nothing but attention, and it is a prerequisite for everything an agency would do afterward.

The second case is a business whose customers do not arrive this way. If your work comes from referrals, partners or outbound, then search is a long, slow channel to build and your money is better spent making the channel that already works less manual. That is a genuine answer, and an agency that never gives it is not advising.

The third case is timing. If your site is about to be rebuilt, most optimization done first gets discarded. Sequence matters more than speed here.

Where this leaves you

The category is real, the change driving it is real, and the pricing is unsettled enough that the same brief can cost very different amounts depending on who you ask. That is the normal condition of a market three years into a shift, and it means the burden of telling the two jobs apart falls on the buyer.

So the single question worth opening with is the one at the top of this piece: which of the two are you selling me, and what will you show me at ninety days? The answer sorts the field quickly.

If you would rather work out whether any of this is your constraint before you go shopping, that is the kind of thing we look at in a consultation — including the outcome where the honest answer is that search is not your bottleneck and your budget belongs somewhere else.