Generative Engine Optimization Agency: What to Buy and Skip.

A generative engine optimization agency helps your business get named, cited and linked when someone asks an AI assistant a question you want to win. The good ones do a short list of unglamorous things: make your entity unambiguous, let the right crawlers in, structure pages so a model can lift an answer from them, earn mentions on the sources models already trust, and measure the result. The bad ones sell placement guarantees and prompt tricks. This guide shows you the difference before you sign anything.

We run this work on our own properties, so the opinions below come from dashboards, not a pitch deck. Our study of AI assistant referrals across 31 websites put AI assistants at 10.1% of all sessions. That is real demand, and it is why this market suddenly has a naming problem and a crowd of new vendors.

GEO, AEO and AI search optimization: three names, one job.

The three labels describe the same work from different angles. Generative engine optimization (GEO) came from the research side and focuses on getting your content used inside a generated answer. Answer engine optimization (AEO) grew out of featured snippet and voice search work and focuses on answering a question cleanly enough to be quoted. AI search optimization is the plain-English umbrella that buyers actually type. Agencies pick whichever label their buyers search for, then ship roughly the same deliverables.

The overlap is near total. All three need the same technical foundation (crawlable pages, clean structured data, one consistent entity), the same content shape (direct answers, specifics, sources), and the same authority signals (mentions and citations on sites the models already lean on). The differences are emphasis, not method. If a proposal charges separately for GEO and AEO, ask which page gets touched twice.

What a generative engine optimization agency actually does.

Entity clarity. A model has to know who you are before it can recommend you. That means one consistent name, description, location and category across your site, your Google Business Profile, LinkedIn, directories, review platforms and press. An agency audits every place your brand is described, fixes the contradictions, and writes an About page and Organization schema that agree with each other. Boring, and the highest use item on this list.

Crawler access. AI platforms fetch pages with their own named crawlers, and each publishes its user agent. Plenty of sites block them by accident through a CDN bot rule, a firewall preset, or a robots.txt line copied from a template. The agency reads server logs, robots.txt and firewall rules, decides deliberately what to allow, and publishes an llms.txt file. That file is a plain markdown index of your most useful pages, proposed by Jeremy Howard in September 2024 and documented at llmstxt.org. No platform is obliged to read it and it takes about an hour, so do it, and do not pay a retainer for it.

Citation-ready pages. Models lift answers from pages that state the answer early, give specifics (numbers, names, dates, prices, steps) and show where the facts came from. The agency rewrites your key pages so the first paragraph answers the question, adds comparison tables, and replaces generic copy with original data or first-hand experience. Pages written years ago to rank on vague keywords usually need this most.

FAQ and schema. A real FAQ block, written in the words customers use, hands a model clean question and answer pairs. Schema (Organization, Service, Product, FAQPage, Article with a named author) removes ambiguity about what a page is and who stands behind it. Neither is magic. Both make extraction cheaper, and models prefer cheap.

Authority. Assistants lean on sources they already cite: review and comparison sites, industry publications, forums, and well-maintained listings. In our 31-site data, review and comparison sites pulled the largest AI share of any site type we own. The agency finds which sources the models cite for your questions, then earns mentions there through listings, contributed data, interviews, reviews, and the occasional original study other people want to quote.

Measurement. If the agency cannot show you referral sessions, citations and share of voice by month, with dates, it is selling theater. The measurement section below covers what that should look like.

What is snake oil.

• Guaranteed placement. Nobody controls what a model says. Anyone promising a top spot in ChatGPT or an AI Overview is promising something they cannot deliver, and they know it.
• AI keyword stuffing. Repeating "best X for Y" forty times does nothing for a language model and makes the page worse for the human who has to read it.
• Prompt gaming. Asking an assistant about your brand a thousand times to "train" it changes nothing about its training data or its retrieval. It changes your invoice.
• Hidden text for bots. Serving one page to crawlers and another to people is cloaking. It has been a Google penalty for two decades and it is not safer because the reader is a model.
• Fake reviews and seeded forum posts. Platforms detect this, and the reputational downside outlives the engagement.
• A separate GEO retainer that duplicates your SEO retainer. Same pages, same schema, same authority work, billed twice.

How to evaluate a GEO or AEO agency.

Treat it like hiring an SEO firm in 2012: young market, lots of confidence, few people who can show their work. These questions separate operators from salespeople, and the answers should be specific enough to write down.

• Which prompts will you track for us, how did you pick them, and how often will you run them?
• Show me a before and after for one engagement: referral sessions, citations and share of voice, with dates on the chart.
• Which AI crawlers do you recommend we allow, which do you block, and why?
• Which third-party sources do the models cite for our category today, and how do you plan to get us onto them?
• What changes on our site in month one, and what do you deliberately leave alone?
• How is this reported alongside our SEO work, and who owns the content calendar?

Proof to ask for is simple. Screenshots of a GA4 AI Assistant channel growing over time. A citation log showing where a client's pages appeared and for which prompts. A monthly prompt panel with dates and the competitors named instead. Case studies that only say "AI visibility up 300%" with no definition of visibility are marketing. Ask what the score counts and who built it.

How agencies measure AI search.

Referral traffic in GA4. Google added an AI Assistant channel to the default channel group on May 13, 2026, according to Google's channel group documentation. Sessions whose referrer matches a recognized assistant get the medium ai-assistant and land in that channel with no setup. It undercounts, because in-app browsers and pasted links arrive with no referrer and fall into Direct, but it is the baseline every agency should be able to show you on day one.

Citations and share of voice. Tools such as Ahrefs Brand Radar run a set of prompts on a schedule across AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini and Copilot, then report mentions, citations and AI share of voice against competitors, as described on Ahrefs' product page. Semrush sells a comparable AI Visibility toolkit. These are useful, and they are exactly as good as the prompt set someone chose on day one.

Manual prompt panels. The cheapest and most honest check: a spreadsheet of 20 to 40 real customer questions, run by hand in each assistant once a month, logging whether you were mentioned, cited or absent, and who got named instead. An agency that does not do this is reading a dashboard instead of reading the answers.

Then connect it to money. On our own agency site, 90 days of lead data showed visitors from ChatGPT converting at 5.2% against 2.4% from Google. The sample was small, and small is the honest word. The direction is the point. Ask your agency to report AI leads, not just AI sessions.

Pricing models, in general terms.

Agencies sell this four ways. A one-time audit (crawler access, entity consistency, citation gap, page readiness) priced as a project. A monthly retainer that bundles fixes, content and reporting, usually tiered by how many prompts are tracked and how many pages or assets ship each month. A monitoring-only subscription that resells a tracking tool with a human summary on top. And a bundled SEO plus AI search retainer, which is the model we think is right, because the work overlaps almost completely.

What should move the price is scope: prompts and markets tracked, pages rewritten, how much third-party outreach is included, how many competitors are benchmarked. What should not move the price is the label. The right structure is one SEO engagement with AI search scope written into it, not two invoices for the same pages.

What usually goes wrong.

• The agency tracks prompts nobody asks: fifty vanity prompts about the brand name, zero about the problems customers actually type.
• Crawler blocks stay in place. A CDN rule from 2023 quietly blocks every AI fetcher, and nobody reads the logs.
• Content gets rewritten for bots and loses the human reader. Citation count rises while conversion rate falls.
• Schema says one thing and the page says another. Contradictions make a source cheaper to skip.
• Reporting stops at sessions. No lead or revenue tie-in, so leadership cannot tell whether any of it worked.
• Authority work gets skipped because it is slow. Mentions on sources models already trust take months, and they are the part that compounds.

Where the agency stops and the work begins.

An agency can audit, fix, write, pitch and measure. It cannot invent the expertise, the original data, the customer proof or the opinions that make a page worth citing. Those come from you: the pricing you are willing to publish, the numbers you are willing to share, the questions your sales team hears every week. The engagements that work pair an operator who knows the mechanics with a client who will put real substance on the page.

Our AI search optimization services are built that way, inside the SEO work rather than next to it. If you want it run properly, Hire Theory Road. If you would rather start yourself, our guide on how to get found by AI search covers the first month of work in order.