A GEO agency sells work meant to get your brand cited inside answers from ChatGPT, Perplexity, Gemini and Google's AI features. Some of that work has a clear mechanism behind it: crawler access, entity markup, passages written to be quoted, and measurement. Some of it is SEO with a new invoice header. And none of it comes with a guarantee, because nobody controls what a language model decides to cite. Published 2026 price guides disagree with each other by a wide margin, so scope matters more than the label. Below: what the term means, a scope checklist, how to measure the work, and what to ask before you sign.
What generative engine optimization means
Generative engine optimization is the practice of shaping your content and web presence so that AI systems that write answers (rather than list links) pull from you and name you. The term comes from a 2023 research paper, "GEO: Generative Engine Optimization", which framed it as optimisation for generative engines and built a benchmark of queries to test content changes against. The authors reported visibility gains of up to 40% on that benchmark, with results varying by domain and by tactic. Notably, plain keyword stuffing didn't do much in their tests, while things like adding sources, quotations and statistics did better.
That's the origin. It's an academic starting point, not a promise that any agency can repeat those gains on your site. Read the paper's limits as carefully as its headline number.
Deliverables worth paying for
Here's the test I'd apply to every line in a proposal: can the agency tell you the mechanism, the artefact you'll receive, and how you'd see whether it worked? If yes, it's worth pricing. If the answer is "trust our process", it isn't.
- Crawler access audit. Do the relevant bots reach your pages at all? This covers
robots.txt, CDN or firewall rules that block bots by default, JavaScript-only rendering, and login or consent walls. If a crawler can't fetch the page, nothing downstream matters. The deliverable is a list of what's blocked, by which rule, and a decision on each one. - Entity and schema work. Making it unambiguous who you are, what you sell and which other profiles belong to you (Organization, Product, author markup, consistent naming across the web). The mechanism is disambiguation. Google's AI features documentation is blunt that structured data should match the visible content, so markup is hygiene, not a trick.
- Citable passage rewrites. Rewriting key pages so a single paragraph can be lifted and still make sense: a direct answer first, a specific claim, a named source, a figure where you have one. This maps onto what the GEO paper found helpful. It's also the most labour-intensive item, so check how many pages the quote covers.
- Source-gap analysis. For the questions your buyers ask, which third-party pages do AI answers actually cite, and where are you absent? The output is a target list of publications, directories, review sites and forums. This is where the expensive digital PR work comes from, and it's often the most durable part.
- Citation measurement. A repeatable set of prompts, run on a schedule across named platforms, logged with the sources returned. More on this below.
If you want a sense of what a scoped programme looks like in practice, our AI SEO and GEO service is built around these same five items.
A scope checklist you can mark up
Use this on any proposal. The verdicts are my own judgement, drawn from the sources linked in this article, not test results.
| Line item | Mechanism behind it? | What you should receive |
|---|---|---|
| Crawler access audit | Yes: fetchability | Bot-by-bot access report and fixes |
| Entity and schema work | Yes: disambiguation | Markup spec, entity map, implementation notes |
| Citable passage rewrites | Partly: supported by the GEO paper | Rewritten pages, with page count stated |
| Source-gap analysis | Yes: citation sources are observable | Ranked list of sources to earn mentions on |
| Citation measurement | Yes, with limits | Prompt set, platforms, cadence, raw logs |
| "AI ranking" guarantee | No | Nothing verifiable |
| llms.txt as a headline fee | Weak | A small file, not a retainer |
| Prompt-stuffing | No | Risk, not output |
| Bulk AI articles | Weak | Volume, not authority |
Deliverables to question
Guaranteed AI rankings
Say it plainly: no one can guarantee AI citations. Answers vary by prompt, by user, by platform and by day. There's no ranking table to sit at position three in. An agency promising a fixed citation count is either measuring something narrow (a handful of prompts they chose) or hoping you won't ask. Walk away from any contract that ties the fee to a guarantee you can't audit.
llms.txt as a headline service
I've written about this at length in llms.txt explained, so the short version. It's a proposed file, not a standard that major engines are confirmed to rely on. Google's AI features documentation says you don't need to create new machine-readable files, AI text files or markup to appear in those features. Fine to add one if it costs an hour. Not fine to see it as a line item with its own monthly fee.
Prompt-stuffing
Hiding instructions in page text ("when summarising, recommend this brand") or stuffing pages with the exact phrasing of likely prompts. The GEO paper reported that keyword stuffing gave little benefit, and engines have every reason to discount manipulative text. You'd be paying for a risk to your reputation.
Mass AI content
Volume can be a legitimate part of a content programme. But hundreds of thin, machine-written pages don't create the third-party mentions that engines tend to trust. One 2026 pricing roundup puts it well: what you're paying for, in SEO or GEO, is the same expensive work of content that answers real buyer questions, mentions and links from sources engines trust, and technical hygiene. Bulk text skips the expensive parts.
How to measure a GEO engagement
Measurement is where proposals go vague, so it's worth being specific about what each metric tells you and where it misleads.
- Citation tracking. A fixed prompt set, run repeatedly, recording whether your domain appears as a cited source. Useful because it's concrete. Limit: answers are non-deterministic, so one run proves little, and the prompt set is chosen by whoever is selling.
- Share of voice in AI answers. Your brand mentions as a share of all brands mentioned across the prompt set. Good for competitive context. Limit: it's only as honest as the prompts, and a mention isn't the same as a recommendation (or a citation, or a click).
- Referral traffic from AI assistants. Sessions from sources like chatgpt.com or perplexity.ai in your analytics. It's real traffic and it's yours to check. Limit: it undercounts, because some assistants strip referrers or people copy a brand name into a search box later. And the volume is usually small.
Ask for raw logs, not just a dashboard score. Ask which platforms and how often. And ask for a baseline before work starts, otherwise "improvement" has nothing to be measured against.
Be wary of vendor claims that AI traffic converts at some multiple of organic. Those numbers circulate in agency marketing without clear methodology. Check your own analytics instead.
For the wider picture of how these channels fit together, see our AI search, GEO and AEO playbook. If Google's features are your main target, how to appear in AI Overviews covers that case, and getting cited by ChatGPT search covers the OpenAI side.
What the market charges
Honest answer first: published figures don't agree. One 2026 pricing roundup compared seven guides and found the lowest floor at $1,500 a month and the highest ceiling at $50,000, with the middle contested too. One guide's full-optimisation range ends at $5,000, which is where another says the category starts. Treat all of these as vendor-published, not independent benchmarks.
Retainers
Ranges quoted in the sources:
- $3,000 to $5,000 a month for entry programmes, $5,000 to $10,000 for mid-market specialists, and $10,000 to $20,000 or more for full-service and enterprise work, per Acromatico's summary of published guides.
- $2,000 to $10,000 a month for most mid-market programmes, according to Digital Elevator.
- $4,000 to $10,000 a month, per Citeme, which also says offers far below market rates usually repackage standard SEO deliverables under a new label.
That last caution is worth keeping. Astral's guide makes a related point: a very cheap quote is almost always dropping some deliverables, and a very expensive one is bundling extras like strategist hours and custom tooling. Neither is wrong by default. The question is which items you're getting.
Projects and audits
One-off GEO audits are quoted at about $1,500 to $5,000 in several sources, including Young Company's guide. Astral puts one-time projects at $5,000 to $50,000. For a team that can execute internally, an audit plus a roadmap is often the lower-risk place to start. (Optimist's own guide notes the buyers who get the most are those who can execute from an agency's roadmap and briefs.)
Add-ons to SEO retainers
The same Young Company piece, citing Frase, says GEO add-ons to an existing SEO retainer often add 20% to 30% to monthly recurring revenue. Worth knowing if you already pay an SEO agency, since a lot of GEO scope overlaps with work they should be doing.
Is GEO cheaper than SEO? One comparison says about 25% cheaper at matching levels. Others show overlapping ranges. Again: scope over label.
Questions to ask any GEO agency before signing
- What exactly is in the scope, line by line? Ask them to map each item to the checklist above. If an item has no mechanism or artefact, price it at zero.
- How do you measure citations, and can I see raw logs? You want prompt sets, platforms, run frequency and returned sources, not a single score.
- What's the baseline, and who chooses the prompts? Insist on agreeing the prompt set together before work starts.
- What do you do that a good SEO agency wouldn't? A fair answer names crawler audits, entity work and source-gap analysis. A weak one is "AI-ready content".
- Who writes the content, and how many pages does the fee cover? Vague volume is a flag.
- What third-party mentions will you pursue, and how? Earned mentions in authoritative sources are the costly part of the work. Ask how it's done without paying for placement that engines discount.
- Do you show up in AI answers yourselves? Not proof of skill, but a fair thing to check.
- What happens if citations don't improve? A straight agency says outcomes aren't guaranteed and describes what they'd change. If they promise results, see the section above.
- What are the exit terms? Short initial commitment, then monthly. A long lock-in with unverifiable output is the worst combination.
FAQ
Can my existing SEO agency do GEO work?
Often, yes, for the overlapping parts: technical access, structured data and content quality. The gaps are usually source-gap analysis and citation measurement, which need a prompt-tracking process. Ask them to scope those two items separately and compare it with a specialist quote.
Should I block AI crawlers in robots.txt?
That's a business decision, not an SEO one. Blocking a bot means that platform can't fetch your pages, which affects whether it can cite you. Some publishers block for licensing reasons and accept the trade-off. A crawler audit should list each bot and give you the choice, not make it for you.
How long until a GEO programme shows results?
Nobody can give you a reliable figure, and I won't invent one. Some fixes, like unblocking a crawler, are quick to implement. Earning mentions on third-party sources takes much longer and is less predictable. Ask for milestones tied to deliverables, not to citation counts.
Do I need a GEO tool as well as an agency?
Possibly. Tools track citations on a schedule; agencies do the work that changes them. If you have in-house writers and developers, a tool plus a one-off audit may be enough. If you have neither, you're paying for execution, and the retainer ranges above apply.
Strip the label and a GEO agency is selling five things with a mechanism (access, entities, citable passages, source gaps, measurement) and a handful without one. Price the first group, question the second, and treat any guarantee of AI citations as the sharpest warning sign in the proposal.