AI SEO / GEO

How to Choose an AI SEO Agency (Without Getting Sold a Rebrand)

Joel House, Founder, Xpand Digital
Joel HouseForbes Agency Council
Founder, Xpand DigitalJuly 11, 202615 min read

Almost every SEO agency added “AI SEO” to its homepage in the last twelve months. Almost none of them changed what they actually do.

That is the real problem you're trying to solve when you set out to hire an AI SEO agency — also sold as GEO (generative engine optimization) or AEO (answer engine optimization). The category is eighteen months old. The demand is real: people are searching in ChatGPT, Perplexity, Gemini, and Google's AI Overviews instead of scrolling ten blue links, and businesses can feel the referral traffic shifting. So agencies slapped a new headline on the same retainer. The pitch deck says “AI-first.” The deliverables are the same monthly blog posts and the same backlink report they were selling in 2021.

I'm writing this from the operator's chair, not the buyer's. I run Xpand Digital's AI SEO practice, I wrote a book called AI for Revenue, and I built the software my own team uses to measure AI visibility. That gives me an unfair view of how this market talks versus what it can prove. This guide is the filter I'd hand a friend who asked, “how do I tell the operators from the opportunists?” It's a set of questions, green flags, red flags, and a cost reality — not a definition of GEO. If you want the “what is it” explainer, that's a different article. This one assumes you've decided you need help and you're trying to pick well.

Why the choice matters more than it looks

The gap between the best and worst AI SEO agency isn't a matter of degree right now — it's binary. Most of the market can't even measure the thing they're selling. According to the MentionLayer Q1 2026 AI Visibility Index— a study of 1,004 businesses across 95,392 data points — 65.9% of businesses are effectively invisible in AI search: they never surface when a model answers a buying-intent question in their category. If two-thirds of businesses are invisible, then “we do AI SEO” is worth exactly nothing unless the agency can show you where you sit on that curve and move you up it.

Here's the asymmetry that makes the hire high-stakes: AI search is a winner-take-most surface. A traditional Google results page shows ten organic listings, a map pack, and ads — lots of slots. When ChatGPT answers “who's the best commercial roofer in Denver,” it names two or three companies and stops. Being cited is a different game from ranking #7. You're either in the answer or you don't exist. The agency you pick is placing a bet on how models choose who to name, and if they're guessing, you find out six months and $30,000 later. That's why the selection criteria below lean so hard on proof of measurement. In a market this young, the ability to instrument the problem is the whole ballgame.

The one question that filters most of the market

Before the ten-point checklist, there's a single question that eliminates the majority of candidates in one move:

“How do you measure whether ChatGPT, Perplexity, and Gemini recommend a business — and can you show me a live example?”

An agency that has actually done this work has an answer that involves running your brand and your competitors through the models on a schedule, tracking which prompts surface you, and quantifying the share of answers you appear in. An agency that hasn't will wave at “optimizing your content for AI” and change the subject. There is no faking the measurement answer, because the tooling either exists in their workflow or it doesn't. Everything in the checklist below is downstream of this: if they can't measure it, they can't improve it, and they definitely can't report on it.

I'll be direct about my own bias here: my team built MentionLayer precisely because the measurement layer didn't exist as an agency-grade product when we needed it. You do not have to hire an agency that owns its software. But you should hire one that runsreal instrumentation, whether they built it or licensed it — because the alternative is paying for a strategy nobody can score.

10 criteria for choosing an AI SEO agency

Run every shortlisted agency through these ten. You don't need a perfect score — you need honesty on each one and no hard red flags. I've written the exact question to ask, what a good answer sounds like, and what should make you close the tab.

01

They can measure AI visibility, not just talk about it

This is criterion one because it's the foundation everything else stands on. AI SEO without measurement is astrology with a Slack channel. The agency should be able to tell you, today, what percentage of relevant AI answers name you versus your competitors — and re-check it every week as models update.

The question to ask
“Walk me through how you'd baseline my AI visibility in week one. What tool produces that number, and how often does it refresh?”
Green flag
They describe a repeatable process: a defined prompt set, tracked across ChatGPT / Perplexity / Gemini / AI Overviews, producing a share-of-answers metric with week-over-week deltas.
Red flag
“We optimize your content so AI understands it better.” No number, no baseline, no tooling named. Vibes as a service.
Joel's take
If an agency can't baseline you before they pitch you, they're selling activity, not outcomes. A real one will often show you your own invisibility on the first call — that's the hook, and it's an honest one.
02

They understand SEO and GEO are different disciplines

Ranking on Google and getting cited by a language model overlap, but they are not the same job. Classic SEO optimizes a page to win a slot on a results page. GEO optimizes an entity— your brand, its facts, its citations across the web — so a model reaches for you when it assembles an answer. An agency that treats “AI SEO” as regular SEO with a new coat of paint will pour your budget into the wrong surface.

The question to ask
“What do you do for AI visibility that you would not do for a normal SEO campaign?”
Green flag
They can name specific AI-era moves: entity consistency across the web, structured data for machine-readability, earning citations on the sources models trust, and monitoring how you're described — not just where you rank.
Red flag
The AI SEO deliverables list is indistinguishable from the 2021 SEO deliverables list: more blog posts, more keywords, more backlinks, done.
Joel's take
The honest agencies will tell you that you often need both — classic AI SEO services and traditional organic still feed each other. What you're screening for is whether they understand the difference, not whether they abandon one for the other.
03

They own or run their own instrumentation

An agency selling GEO without its own measurement stack is guessing with your money. The tooling to track model citations, run prompt panels at scale, and detect when an update changes who gets named is not something you improvise in a Google Sheet. Either they built it, or they licensed a serious platform and use it daily. Ask which.

The question to ask
“What software runs your AI visibility tracking — and is it yours, licensed, or a spreadsheet?”
Green flag
A named platform they operate consistently, ideally with their own product or original research behind it. Agencies that have published data on AI search have skin in the game and a reason to be right.
Red flag
They manually paste prompts into ChatGPT once a month and screenshot the results. That's a demo, not a measurement system, and it won't survive a model update.
Joel's take
This is the differentiator I hold my own team to. We built MentionLayer and the PressForge digital-PR engine because you cannot improve what you cannot measure, and you cannot earn citations at scale by hand. You don't need to hire us — but hold everyone to that bar.
04

They have a real citation-earning engine, not just on-page tweaks

Models cite sources they trust. A huge share of AI visibility comes from being mentioned on the third-party sites, publications, directories, and reference pages that language models weight heavily. That means the agency needs a way to earn citations off your own site— digital PR, data studies, expert placements — not just tidy your homepage schema. On-page work is table stakes; the citation engine is the moat.

The question to ask
“How do you get my business mentioned on the sources these models actually trust? Show me a placement you earned for a client.”
Green flag
A repeatable digital-PR or citation-building motion with examples: earned mentions on industry publications, original research that gets referenced, structured presence on the directories models read.
Red flag
The entire program lives on your own website. If nothing they do happens off-domain, they're optimizing a signal that models already discount.
Joel's take
This is where most “AI SEO” offerings quietly collapse. Earning citations is hard, slow, relationship-driven work — the same digital-PR muscle that built durable link profiles for a decade. Agencies that already had that muscle adapt fast. Agencies that only ever did on-page can't fake it.
05

Their proof is real and checkable

The AI SEO space is new enough that fabricated case studies are everywhere — screenshots with no client name, “340% more AI citations” with no methodology, testimonials from businesses that don't appear to exist. Because so few buyers know how to verify AI-visibility claims, the temptation to invent them is high. Your job is to make the proof concrete.

The question to ask
“Can I see a named case study, and can I contact that client? What exactly did you measure, before and after?”
Green flag
Named clients, checkable results, a clear before/after on a defined metric, and a willingness to say “here's one that didn't work and why.” Real operators have losses.
Red flag
Anonymous screenshots, round-number miracle stats, and stock-photo testimonials. If you can't verify it, price it at zero.
Joel's take
I'm militant about this because I've watched fabricated proof become the default in young categories. Ask for the client's phone number. The agencies that flinch are telling you everything. The ones that hand it over just won the pitch.
06

They optimize entities and structure, not just keywords

Language models don't rank keywords — they assemble answers from what they understand about entities and their relationships. That makes machine-readability the technical core of GEO: consistent business facts across the web, structured data that spells out who you are and what you do, and content written to be extracted and quoted rather than skimmed. An agency stuck in keyword-density thinking is optimizing for an algorithm that's being replaced.

The question to ask
“How do you make my business machine-readable? What do you do with structured data and entity consistency?”
Green flag
They talk about schema markup, entity consistency across directories and profiles, clear factual content, and writing in a way models can quote cleanly. They think in facts, not keywords.
Red flag
Everything comes back to keyword volume and density. That's a 2015 mental model wearing a 2026 label.
Joel's take
Structured, factual, quotable content is quietly the highest-leverage on-page work in GEO — and it's the part most content-mill agencies skip because it's unglamorous. It's also what makes your pages eligible to be the sentence a model reads aloud.
07

Their reporting is about outcomes, not activity

Ask to see a real monthly report from an existing client (redacted is fine). This single artifact tells you more than the whole sales call. A report full of “12 blog posts published, 40 keywords tracked, 8 backlinks built” is an activity log. A report that leads with “your share of AI answers went from 4% to 19% on these buying-intent prompts” is an outcome. You're paying for the second kind.

The question to ask
“Show me a sample monthly report. What's the headline number you hold yourselves to?”
Green flag
Reporting anchored to AI visibility and citation share, tied where possible to traffic and leads, with honest commentary on what moved and what didn't.
Red flag
Reports that count outputs (posts, links, hours) instead of results, or that hide behind “AI search is impossible to measure” — which we've already established is false.
Joel's take
Activity reports exist to make you feel like you got your money's worth without proving you did. The tell is whether the first page of the report is a number that maps to your revenue or a tally of things they did.
08

They have a plan for when the models change

The models update constantly. A prompt that named you in March might not in June because OpenAI shipped a new version or changed how it weights sources. An agency without a monitoring-and-response loop will only discover you dropped out of answers when you notice your leads dried up. You want a team that watches for volatility and adjusts — not one that delivers a static “GEO audit” and disappears.

The question to ask
“A model update drops my citations next quarter. What's your process for catching it and responding?”
Green flag
Continuous monitoring, alerting on visibility drops, and a stated cadence for re-testing and adapting. They treat GEO as an ongoing position, not a one-time project.
Red flag
They sell a fixed-scope “AI SEO audit” as the whole engagement, with no monitoring after delivery. You'll be flying blind the moment the models shift.
Joel's take
This is the difference between hiring a mechanic and buying a map. The surface moves under you every few weeks. Pay for the ongoing hand on the wheel, or don't bother.
09

Pricing and contract terms are transparent

A serious AI SEO engagement is real work — measurement, content, digital PR, technical structure, monitoring — and it's priced accordingly. Be equally suspicious of the $500/month “AI SEO package” (nobody earns citations at that price) and the vague enterprise quote with a twelve-month lock-in and no exit. Transparent pricing tied to a clear scope is a proxy for how they'll treat you for the whole relationship.

The question to ask
“What does this cost, what exactly do I get for it, and what are the contract terms if it isn't working?”
Green flag
Clear scope, clear price, month-to-month or short initial terms, and a straight answer on what happens if you leave. Confidence shows up as flexibility.
Red flag
Long lock-ins with punitive exits, prices that only appear after three calls, or a suspiciously cheap package that can't possibly fund real citation work.
Joel's take
I run Xpand month-to-month on purpose. If the work is good, you stay because you want to. Long contracts exist to protect agencies from the consequences of mediocre work — that's the whole reason they push them.
10

They'll tell you when you don't need them

The last filter is character. A good agency will sometimes tell you that AI SEO isn't your bottleneck yet — that your bigger problem is a broken conversion path, thin traditional SEO, or a market where AI referral volume is still tiny. That honesty costs them a sale and earns them a reputation. It's also the single best predictor of whether they'll tell you the truth once you're a paying client and the news is bad.

The question to ask
“Is AI SEO actually the right priority for my business right now — or is something else more urgent?”
Green flag
They diagnose before they prescribe, and they're willing to say “not yet” or “fix this first.” They'd rather be right than booked.
Red flag
Every business, in every situation, urgently needs their exact package this quarter. A hammer that sees only nails.
Joel's take
You're not just buying a service, you're buying who tells you the truth for the next year. Hire the one who was willing to talk you out of the sale — they're the one who'll tell you when a campaign is underperforming instead of hiding it in an activity report.

“Agencies selling GEO without their own instrumentation are guessing. In a market where two-thirds of businesses are invisible, guessing is the most expensive thing you can buy.”

Joel House · Founder, Xpand Digital

Seven claims that should make you walk

If you hear any of these in a sales call, treat it as disqualifying until proven otherwise. Each one is a marker of an agency that bolted “AI” onto an old offer without changing the work underneath.

  • “We guarantee you’ll rank #1 in ChatGPT.”
    There is no #1 in a generative answer, and nobody controls what a model says. A guarantee like this is either a misunderstanding of how the surface works or a deliberate lie. Both should end the conversation.
  • “AI SEO is basically the same as regular SEO.”
    Then why are you paying an AI-SEO premium for it? This line means their deliverables didn’t change — they’re selling entity-and-citation work they don’t actually do.
  • “We can’t really measure AI visibility — nobody can.”
    False, and disqualifying. Measurement platforms exist and produce weekly share-of-answer numbers. An agency saying this is telling you they don’t have the tooling, and asking you to fund a program with no scoreboard.
  • “Just give us six months and you’ll see results.”
    You should see a visibility baseline in week one and directional movement well before month six. ‘Trust me for half a year’ with no interim metric is how you lose $30,000 quietly.
  • “Here are our results — but we can’t share client names.”
    Some confidentiality is legitimate, but a whole portfolio of anonymous miracle stats is not. If nothing is checkable, assume nothing is true.
  • “We’ll flood your site with AI-written content.”
    Volume for its own sake is the opposite of GEO. Models reward clear, factual, quotable, trustworthy content — and Google’s helpful-content systems suppress the mass-produced kind. This tactic actively works against you.
  • “Sign here — it’s a standard twelve-month agreement.”
    In a discipline this new, where the models shift monthly, a long lock-in transfers all the risk to you. Confident operators earn renewals; they don’t trap them.

What it actually costs

Pricing in AI SEO is still settling, but the honest ranges track the real work involved. Use these as sanity checks, not quotes — your market, competition, and starting position move the number.

  • Under $1,000/month: almost always on-page tweaks and AI-written content with no citation engine and no measurement. For most businesses this is money spent to feel busy.
  • $2,000–$5,000/month: the realistic floor for a program that includes measurement, structured on-page work, and a genuine citation/digital-PR motion. This is where most $500K–$5M businesses should land.
  • $5,000–$15,000+/month: competitive markets, national footprints, or aggressive timelines where the citation and PR work has to move fast against entrenched competitors.
  • Project / audit engagements: a one-time AI visibility audit is a reasonable way to start — get the baseline and the gap analysis before you commit to a retainer.

The single most useful thing you can do before signing anything is to get an independent baseline. Know your starting share of AI answers, and you can hold any agency accountable to moving it — which is the entire point of hiring one.

If you're still building your shortlist, a few resources worth reading next:

Frequently Asked Questions

Joel House, Founder, Xpand Digital
Founder, Xpand DigitalJuly 11, 2026

Want to see where you
actually stand in AI search?

We'll baseline how often ChatGPT, Perplexity, Gemini, and AI Overviews name your business versus your competitors — the same measurement you should demand from any agency. No pitch deck, just the number.

Get an AI Visibility Audit How we run AI SEO
Put it to work

Work with the team that wrote the playbook.