All Posts
Generative Engine Optimization

What Does a GEO Agency Actually Do? An Operator's Breakdown

Joel House, Founder, Xpand Digital
Joel HouseForbes Agency Council
Founder, Xpand DigitalJuly 11, 202614 min read

Most people asking this question are trying to work out one thing: is a GEO agency real work, or a rebranded SEO upsell with a new acronym?

Fair question. “Generative engine optimization” became a pitch deck term almost overnight, and plenty of agencies bolted it onto a retainer without changing a single deliverable. So instead of another “what is GEO” explainer, this is the operator's version: I run a GEO agency, and here is the actual work that happens between the contract and the result — broken into the seven workstreams a real program runs, what each one produces, and how you'd verify it was done rather than invoiced.

The short definition, so we're on the same page: a GEO agency gets your business named and cited inside the answers AI engines give — ChatGPT, Perplexity, Gemini, and Google's AI Overviews. Not ranked in a list of blue links a human scrolls through. Cited inside the synthesized answer, where more and more buying decisions now start and, increasingly, end. That's a different target than classic SEO, and it changes what the day-to-day work looks like.

It matters because the gap is enormous. The MentionLayer AI Visibility Index— an industry study across roughly 1,000 businesses — found that about two of every three are effectively invisible in AI search. Most companies aren't losing the AI-answer race; they haven't entered it. A GEO agency's entire job is to change that, methodically.

Disclosure: we practice this on our own stack. Xpand built MentionLayer, the AI-visibility SaaS that instruments the measurement below, and PressForge, the digital-PR engine that earns the citations. I also wrote a book on it — AI for Revenue. I'll give you the principle first every time; the workstreams below hold whether or not you use any of that.

The 7 Workstreams a Real GEO Program Runs

A GEO engagement isn't one activity — it's seven workstreams that stack. Skip the first (measurement) and everything after it is guesswork. Run them in order and each one compounds the next. Here is what a competent agency or GEO consultant is actually doing inside each.

01

Baseline Your AI Visibility

This is the workstream that separates a real GEO agency from a buzzword seller, and it's the one most skip. Before anything gets optimized, you measure where you stand: across a defined set of the questions your buyers actually ask, how often does each AI engine name you, cite you, or recommend a competitor instead?

What the work actually is
We build a prompt set — typically 30 to 100 buyer questions spanning your category, your named competitors, and your specific services — then run them across ChatGPT, Perplexity, Gemini, and Google AI Overviews on a repeatable schedule. For each answer we record whether you appear, in what position, what the model said about you (accurate or not), and which sources it cited to get there. That last part is the goldmine: the citation sources tell you exactly which properties to influence. Manual spot-checks don't scale, which is why the measurement runs through instrumentation — in our case MentionLayer — so the baseline is a dataset, not a vibe.
What you get
A share-of-answer baseline across all four engines, plus a ranked list of the exact source URLs each model cites for your category.
How you'd verify it
Ask to see the prompt set and the raw baseline before work starts. A real agency shows it; a buzzword seller reports 'rankings' instead.
Joel's take
If an agency can't show you a baseline, walk. Everything else in GEO is unmeasurable without it, and “trust us, it's working” is exactly how the SEO industry earned its reputation. A free version of this baseline is the entire point of our AI visibility audit — the audit is the measurement method.
02

Build the Entity Foundation

Before a model can recommend you, it has to be confident it knows who you are. Entity work is the unglamorous plumbing that makes your business a clean, unambiguous thing the model can identify and describe accurately — and it's where the fastest wins usually hide.

What the work actually is
We reconcile how your business is represented everywhere a model learns about entities: structured data (Organization, Person, Product, and Service schema) on your own site; consistent name, address, and description across your directory and profile footprint; a Wikidata entry and, where it's warranted, a Wikipedia presence; and clean sameAslinks tying your profiles together. The goal is that every engine returns the same accurate set of facts about you — what you do, who you serve, what you're known for — instead of a confident hallucination or a blank.
What you get
Deployed schema, a reconciled profile footprint, and a knowledge-graph presence that returns accurate facts about your business.
How you'd verify it
Ask the engines directly — 'who is [your company] and what do they do?' — before and after. Accuracy and completeness should visibly improve.
Joel's take
This is the highest-return early work and the cheapest to fix. A model that describes you wrong isn't going to recommend you, and a lot of “we're invisible in AI” problems are really “the model doesn't know we exist as a distinct entity.” Fix the identity layer and the rest of the program has something to build on.
03

Seed the Sources Models Actually Cite

Here is the part that looks the most like traditional PR and matters the most for competitive terms. Models don't invent recommendations from nowhere — they synthesize from sources they trust. If you're not in those sources, you don't get cited, no matter how good your own site is.

What the work actually is
Workstream 01 already told us which URLs each model pulls from for your category — the listicles, the industry publications, the review platforms, the forums, the comparison sites. This workstream gets you accurately represented in those specific places through digital PR: earning editorial coverage, getting included in the roundups models quote, contributing expert commentary to the outlets in the citation set, and correcting the third-party pages that describe you wrong. It's the same discipline as earning high-quality backlinks, pointed at a different target — the corpus the model reads, not just the pages that pass PageRank. We run this through PressForge because doing it at volume by hand is brutal.
What you get
Editorial placements and corrected third-party representations in the exact sources your target engines cite for your category.
How you'd verify it
Re-run the baseline prompt set. New citations from newly-seeded sources should start appearing in the answers over the following months.
Joel's take
This is the workstream cheap “GEO” providers quietly skip because it's the hardest and slowest — and it's the one that actually moves competitive answers. If a proposal is all on-site content and no source seeding, it will get you found for your own brand name and nothing else. Real digital PR is the moat, same as it was in SEO. Our link building guide covers the mechanics of earning those placements.
04

Restructure Content for Extraction

Models don't reward the same content shape humans skim. They reward content that's easy to lift a clean, quotable claim from. A GEO agency rebuilds your key pages so a model can extract an answer, attribute it to you, and trust it.

What the work actually is
We restructure priority pages to lead with the answer, then support it: clear question-shaped headings, self-contained factual claims that don't need three paragraphs of context to make sense, specific numbers and dates rather than vague ranges, comparison tables the model can parse, and explicit statements of who something is for. We also publish the genuinely citable assets — original data, benchmarks, definitive how-tos — that give a model a reason to quote you by name. This is the layer where answer engine optimizationand classic on-page SEO overlap most, but the bar is higher: it's not “is this readable,” it's “can a model quote one sentence of this out of context and be correct.”
What you get
Rebuilt priority pages structured for extraction, plus citable original assets a model has reason to quote by name.
How you'd verify it
Check whether the model's answer paraphrases or quotes your restructured page, and whether the facts it pulls are the ones you made easy to extract.
Joel's take
The trap here is treating GEO as “write more blog posts.” Volume is not the lever — extractability and citation-worthiness are. Ten pages a model can confidently quote beat a hundred it skims past. Fewer, sharper, more quotable.
05

Manage Reviews and Corroborating Signals

When a model decides whether to recommend a local or considered-purchase business, it leans on corroboration: do independent sources agree this business is legitimate, well-regarded, and a fit for the query? Reviews and third-party sentiment are a big part of that signal.

What the work actually is
We work the signals that tell a model “other people vouch for this business”: review volume, recency, and rating across the platforms that matter for your category; the substance of what reviews say (models read the text, not just the stars); consistent presence in the directories and association listings your industry trusts; and the sentiment of the third-party pages already talking about you. For a considered purchase, corroboration across independent sources is often the deciding factor between two otherwise similar businesses in an AI answer.
What you get
A stronger, more recent, more descriptive review and third-party-signal profile across the platforms your engines weigh.
How you'd verify it
Watch whether the model starts citing review platforms or directories about you it previously ignored, and whether its description of your reputation improves.
Joel's take
This is where GEO quietly reconnects with real business fundamentals. You can't game your way into being recommended if independent sources don't back you up — which is healthy. The agencies pretending otherwise are the ones you don't want.
06

Fix Crawlability for AI Engines

None of the above matters if the AI crawlers can't read your site. This is the technical hygiene layer — less exciting than the strategy, but a hard prerequisite. If GPTBot, PerplexityBot, and the rest can't access and parse your content, you've optimized for an audience that never arrives.

What the work actually is
We audit and fix how AI crawlers experience your site: correct robots.txt directives for the AI user-agents you want to allow, an llms.txtfile that points them at your most citable content, server-rendered content that doesn't hide behind client-side JavaScript the crawlers won't execute, clean and fast pages, and accessible structured data. It's the same instinct as technical SEO, tuned for a new set of bots with their own quirks.
What you get
An AI-crawler-ready site: correct bot directives, an llms.txt map, server-rendered citable content, and no rendering gaps.
How you'd verify it
Check your server logs for AI-crawler hits and confirm the content those bots fetch is the fully-rendered version, not an empty shell.
Joel's take
Boring and non-negotiable. I've seen businesses spend months on content and PR while their most important pages rendered blank to the exact crawlers they were trying to reach. Do this early so the rest of the program isn't pushing rope.
07

Monitor How Answers Change Over Time

GEO is not a project you finish. Model answers drift — engines update, competitors seed new sources, your citations age. The last workstream is the ongoing one: re-running the measurement and reporting what actually moved.

What the work actually is
We re-run the baseline prompt set on a schedule — monthly for most programs — and track the deltas: share of answer per engine, new and lost citations, changes in how the model describes you, competitor movement, and referral traffic arriving from AI engines. That feedback loop tells us which of the earlier workstreams to double down on and where a competitor is quietly out-seeding you. Without it, you're flying blind between quarterly guesses.
What you get
A recurring report showing share-of-answer movement, citation gains and losses, and the next priorities — tied back to the original baseline.
How you'd verify it
Insist the monthly report compares against the original baseline on the same prompt set. Moving goalposts is how weak agencies hide flat results.
Joel's take
The reporting cadence is where you find out whether you hired an operator or a storyteller. Ask upfront exactly which metrics get reported and how often. If “share of answer” and “citation count” aren't on that list, they're not measuring GEO — they're measuring something adjacent and hoping you don't notice.

“An agency selling GEO without its own way to measure a model's answers is guessing. The measurement isn't part of the work — it is the work.”

Joel House · Founder, Xpand Digital

GEO Consultant vs GEO Agency: Which Do You Need?

“GEO consultant” and “GEO agency” get used interchangeably, but they buy you different things. Understanding the difference stops you overpaying for capacity you already have — or underpaying for execution you can't staff.

A GEO consultantis usually one experienced operator. They run the baseline (workstream 01), diagnose where you stand, and hand you a prioritized roadmap across the other six. You or your team execute. That works well if you already have marketing capacity — a content person, someone who can implement schema, an in-house PR or outreach function. You're buying judgment and direction, not hands.

A GEO agencyruns the execution too — the digital PR, the entity and schema deployment, the content restructuring, the ongoing monitoring. You're buying the whole program delivered. That suits businesses without the in-house bandwidth to act on a roadmap, or that want to move faster than a small internal team can.

The honest test: read the seven workstreams above and ask which ones your team could genuinely own. If it's most of them, a consultant and a good roadmap is the efficient buy. If it's few of them — especially workstream 03, source seeding, which needs real PR muscle — an agency will get you there faster. We package both, and the deciding factor is almost always your in-house execution capacity, not your budget.

How to Spot a GEO Agency That's Faking It

Because GEO is new and demand is running ahead of competence, the market filled with providers selling the acronym without the work behind it. These are the tells. Any one of them should make you ask harder questions; two or more and you're looking at a rebranded retainer.

  • No baseline measurement
    They can't show you where you currently stand across the AI engines because they never measured it. If the proposal jumps straight to deliverables with no share-of-answer baseline, there's no way to prove anything worked. This is the single biggest tell.
  • “We check it manually in ChatGPT”
    Spot-checking a few prompts by hand isn't measurement — it's cherry-picking. Model answers vary by session, location, and phrasing. Without instrumentation running a consistent prompt set on a schedule, the reporting is anecdote dressed as data.
  • All on-site content, no source seeding
    A program that's entirely blog posts and schema with no digital PR will get you cited for your own brand name and nothing competitive. Models synthesize from third-party sources; if the agency isn't earning presence in those sources (workstream 03), they've skipped the hard, decisive part.
  • Rankings and traffic as the only report
    GEO's core metrics are share of answer and citation count. An agency reporting only keyword rankings and sessions is measuring classic SEO and calling it GEO. Those metrics can move for reasons that have nothing to do with your AI-answer presence.
  • Guaranteed placement in AI answers
    No one controls what a model outputs. Anyone guaranteeing you'll appear in ChatGPT for a given query is either naive or lying. What a real agency guarantees is the work and the measurement — not the engine's output, which no vendor owns.
  • No tooling of their own or licensed
    The measurement in workstream 01 and the monitoring in workstream 07 require instrumentation. An agency with no visibility tooling — theirs or a serious platform they license — is running GEO on spreadsheets and guesswork. Ask what they use to track model answers over time.

The through-line: everything hinges on measurement. An agency that measures can prove its work; an agency that can't is asking you to take AI visibility on faith — the exact posture that made buyers distrust SEO for two decades. This is the reason we built our own instrumentation before we sold the service, and it's the differentiator worth pressing every provider on.

The Questions to Ask Before You Hire

You don't need to be technical to separate the operators from the storytellers. Ask these five, in this order, and listen for specifics rather than reassurance:

  • How do you measure my baseline AI visibility, and will you show me the prompt set? A real answer describes a defined prompt set run across multiple engines. A weak one says “we monitor ChatGPT.”
  • What do you use to track model answers over time? Listen for named instrumentation. “We check manually” means it won't scale or hold up in reporting.
  • How do you earn citations in the sources models actually pull from? This surfaces whether they do real digital PR (workstream 03) or just publish on-site content and hope.
  • What exactly do you report, and how often? “Share of answer” and “citation count” should be on the list. Rankings-only reporting is a red flag.
  • What does this cost, and what does each tier buy? A straight answer signals a real service. “It depends” with no ranges signals a shapeless one — we publish ours on the GEO pricing page precisely so you can benchmark.

What a GEO Agency Actually Costs

Pricing tracks the workstreams. A foundation program — baseline, entity work, crawlability, content restructuring, and light monitoring — sits at the lower end. Add competitive source seeding through sustained digital PR (workstream 03), and cost rises the same way an aggressive link-building program does, because it's the same labor pointed at a new target.

In 2026, standalone GEO engagements typically run from around $1,500/month for a small-business foundation up to $8,000+/month for full digital-PR-led programs in competitive markets. One-off audits and consultant roadmaps are usually priced as fixed projects. Rather than hide behind “it depends,” we lay out the real ranges and what each tier buys on the GEO pricing page — and the logic for how we scope an engagement on the GEO agency page.

The value math is straightforward: with roughly two of three businesses invisible in AI search, the cost of doing the work is low relative to the cost of being the company a model never mentions while a competitor gets named in every answer. Early movers are claiming that ground now, the way businesses claimed local search in 2012.

GEO is one part of how a business gets found in 2026. These are the pieces that connect to it:

  • GEO agency— how we run the full generative engine optimization program, with the instrumentation behind it.
  • Answer engine optimization— the sibling discipline focused on winning the direct-answer slot across engines.
  • AI visibility audit— the free baseline measurement (workstream 01) as a standalone starting point.
  • GEO pricing— real 2026 ranges and what each tier of the program buys.
  • How to get found in AI search— the strategy-level companion guide to the operational breakdown above.

Frequently Asked Questions

Joel House, Founder, Xpand Digital
Founder, Xpand DigitalJuly 11, 2026
An operator's desk — a printed prompt set marked up in terracotta ink, a laptop showing a model's answer, and index cards sorting citation sources by engine
On the work

A GEO agency's job isn't to make noise. It's to make a model confident enough to name you — then prove, on the same prompt set, that it did.

Joel House · Xpand Digital

Want to see where you stand
inside the AI answers?

We'll run your baseline across ChatGPT, Perplexity, Gemini, and AI Overviews — the exact measurement from workstream one — and show you which sources the models cite instead of you. No pitch deck, just the data.

Get a Free AI Visibility Audit See How Our GEO Program Works
Put it to work

Work with the team that wrote the playbook.

Where to start