How to choose an AI SEO agency

In the first four months of 2026, 68.01% of US Google searches ended without a click.

Two years earlier that figure was 60.45%.

That is the number every AI SEO agency is now selling against, and almost all of them added the label to the homepage without changing what they do. From the outside the two are indistinguishable, and the pitch deck will not tell you which one you are talking to.

Here is how to tell.

What actually changed in SEO

Traditional search ranked pages against a query. You competed for a position on a list, and the list had ten places on it.

An AI answer has no list. The model retrieves a handful of sources and writes one response. You are either cited in it or you are not mentioned at all.

The clearest evidence that these are now two different systems.

In March 2026, we found that the share of Google AI Overview citations coming from pages that also rank in the top ten fell from 76% to 38%.

ALM Corp

Ranking well used to mean getting cited. Now it means getting cited slightly more often than a coin flip.

The rest of the data says the same thing from different angles. When an AI Overview appears, click-through rate on the top organic result drops by roughly 60% (SparkToro, 2026). Zero-click rate rises to about 83% with an AI Overview present, against roughly 60% without (Similarweb). In Google’s AI Mode it reaches 93% (Semrush, 2025). And Pew Research found only about 1% of AI Overview views result in a click on a cited source.

Read those together and the conclusion is uncomfortable. Being cited is now worth more than ranking, and being cited rarely sends you traffic. We covered the mechanics of this in our piece on AI Overview optimization.

Ask any agency to explain that distinction in their own words. If they cannot do it without reaching for the word algorithm, they have not done this work yet.

In AI SEO, fewer clicks, better clicks

All of that reads as pure downside, and it is not.

The traffic that does arrive from AI answers converts better, because the person has already done their research inside the conversation. They arrive informed rather than curious. Similarweb measured ChatGPT referrals converting at 11.4% against 5.3% for organic search across ecommerce sites in 2025, roughly double.

HubSpot reports a larger effect. Aja Frost, their Senior Director of Global Growth, told TechRadar in July 2026 that the company’s own answer engine work produced a 1,850% increase in qualified leads, and that those leads convert at three times the rate of traditional search leads. Worth noting this is HubSpot’s own data, published while launching an AEO product, so treat the exact multiple with more caution than the direction.

The honest caveat is volume. Microsoft Clarity data from November 2025 put AEO traffic at under 1% of total site traffic. A three times better conversion rate on under 1% of your visitors is real and it is small. Anyone selling this as a replacement for your current acquisition is overstating it.

That is the actual case for doing the work. Not that AI search will replace your traffic, but that the share of it that exists is unusually high quality and growing, and being absent from it costs you the buyers who never reach a results page at all.

Measurement tools and firms that change the outcome for AI SEO

This is the distinction costing buyers the most money right now.

There are good tools that track how often models mention your brand. HubSpot shipped one in April 2026. Profound, Peec and several others do versions of the same thing. They run prompts on a schedule, record who gets named, and chart it over time. That is useful, and it is worth what a software subscription costs.

Some agencies have wrapped a service around exactly that and charge agency fees for it. You receive a monthly report showing your citation rate. The number is real. Nothing in the engagement is designed to move it.

The test takes one question. Ask what they will change, not what they will show you. If every answer is about visibility into the problem rather than work on it, you are buying a dashboard with a retainer attached.

Six questions to ask an AI SEO agency

How will you measure whether we get cited?

What should reassure you: a fixed set of buyer-style prompts run on a schedule, recorded, with competitors tracked alongside you. What should worry you: any answer about traffic or rankings. Those are real metrics measured elsewhere, and with roughly 1% of AI Overview views producing a click, traffic is the wrong proxy for citation.

What is your baseline process?

They should measure before touching anything and show you the starting position. If work begins immediately, nobody can prove later whether any of it worked, which suits the agency more than it suits you. Our post on AI brand visibility covers how to run that baseline yourself in an afternoon.

What changes on our site, and what changes off it?

The answer should cover both, with specifics. Models draw heavily on sources you do not own, and with top-ten pages now supplying only 38% of AI Overview citations, an agency working only on your pages is working on the smaller half of the problem. On-page only, described as content optimisation, is the answer to worry about.

What have you seen work, and over what period?

Look for a specific example with a timeframe measured in months, and an admission of what did not move. A case study where everything improved and nothing was difficult is a warning sign. Apply the same scepticism to vendors quoting their own results, including the ones quoted earlier in this post.

What do you do that traditional SEO does not?

Concrete differences, usually around category language, comparison content that names competitors, and third party presence. If the deliverables are the same ones with different labels, ask for the list and check it against what they sold in 2024.

What would you refuse to do?

The most revealing question on the list. Everyone can describe what they offer. An agency that has thought about this has a real answer. An agency that will do anything you ask has no point of view, and a point of view is what you are paying for.

What nobody can guarantee

Three promises worth walking away from.

Guaranteed citations. Nobody controls what a model retrieves. You can make yourself more likely to be cited. You cannot make it certain.

A fixed position in an answer. There is no position. That phrasing usually means someone is still thinking in rankings and has not updated their model of how this works.

A timeline to results. Models refresh on their own schedules. Honest answers here sound like a range with a caveat, not a date.

None of that means the work does not produce results. It means confident promises about a system nobody controls should make you more suspicious, not less.

When you do not need an agency yet

Two situations where hiring anyone is premature, including us.

You cannot yet say what category you are in, in one sentence, without hedging. If the positioning is not settled, no amount of visibility work helps. A model cannot describe you clearly if you cannot.

What you actually want is volume. If the real goal is more published pages, that is a content brief, and you will get it cheaper from a writer than from a firm charging for strategy on top.

Neither is a permanent no. Both are sequencing problems, and doing them in the wrong order wastes a quarter. If you want to see how we approach the work itself, our SEO and AI visibility page sets it out.

Common questions