The GEO agency playbook: price, deliver, report

What does an agency actually sell when it sells GEO?

An SEO retainer sells position. A GEO retainer sells presence inside the answer. The difference is not cosmetic, because the retrieval mechanics are different.

Google’s AI Mode does not run a query once. Google describes it plainly: AI Mode "uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf" (Google, May 2025). One buyer question becomes a spray of sub-questions, and a page can be cited for a sub-question nobody optimized for.

That has a commercial consequence you can put in a proposal. Ahrefs’ March 2026 analysis of 863,000 keyword SERPs and 4 million AI Overview URLs found that only 37.1% of URLs cited in AI Overviews also rank in the organic top 10. Roughly six in ten citations come from somewhere a rank tracker is not looking.

So the deliverable is coverage of an answer space. Your client’s existing ranking report cannot tell them whether they exist in it. Someone has to measure it, and that someone can be you. If you need the category definition to hand a prospect, we keep one at what GEO is and a side-by-side at how GEO differs from SEO.

Is GEO worth adding as a service line?

Two halves, and you need both to sell it without getting caught out in month three.

The demand half. Google reported AI Overviews at over 2.5 billion monthly active users and AI Mode past one billion monthly users in June 2026. OpenAI reported 900 million weekly active users for ChatGPT in February 2026. The audience is not speculative.

The visitors who do arrive behave well. Semrush found the average AI search visitor "is 4.4 times as valuable as the average visit from traditional organic search, based on conversion rate".

Sell it as what it actually is: brand presence in the surface absorbing the research phase. The Pew Research Center studied 68,879 Google searches by 900 US adults and found users clicked a traditional result in 8% of visits where an AI summary appeared, against 15% where none did, and clicked a link inside the summary in just 1% of visits.

The click is thinning. What survives is being the thing the answer names. That pitch holds up in a boardroom because every number in it is checkable.

What goes into a GEO retainer?

The list is not the product. The loop is: measure, find the gap, publish against it, measure the same prompts again. A retainer that only measures is a dashboard subscription with a markup, and clients work that out fast.

What does the month actually look like?

1. Week 1, measure. Full sweep of the prompt set across all five engines. Google organic rank comes out of the same query call, so the classic ranking report is a byproduct rather than a second workflow.
2. Week 2, diagnose and brief. Read the gaps. Prompts where a competitor is named and your client is not. Prompts where your client is named but described in terms they would hate. Prompts with no clear winner at all, which are the cheap ones to take.
3. Week 3, publish. Articles go into the client’s own CMS: WordPress, Webflow, Shopify, Wix, or a webhook into whatever they actually run.
4. Week 4, re-measure and report. Sweep the same prompt set. Report movement on the specific prompts the content targeted. Some will move, some will not, and saying so is what makes the ones that move believable.

How do you price and package it?

Start from your cost floor. The Agency plan includes 10,000 queries, 1,000 prompts, 50 articles, 10 site audits, 50 Segment Radar runs, 2,000 keywords and 10 webhooks (full plan detail).

The query arithmetic sets your cadence. A query is one prompt on one engine on one run. At the full 1,000 prompts across five engines, one complete sweep costs 5,000 queries, so 10,000 queries buys two full sweeps a month. Run half the prompt set instead and you get four.

Whatever you charge, the platform cost divided across that book is your cost of goods. At 20 clients it is a small fraction of any realistic retainer. Your real cost is the strategist hour in week 2, which is the part worth pricing.

Two packaging notes that keep margin intact. Cap the prompt set per tier, because prompts are the unit that actually consumes budget, and an unbounded set is how a retainer quietly goes underwater. And put content volume in the contract, since content is the lever that moves the measurement, so it is the part clients will ask for more of.

All five engines are in every paid plan with no per-engine add-ons, which means you can quote a client "all five" without a pricing conversation with us first. There is a free plan, so you can pilot a single client before committing the book.

How do you report it to a client?

White-label the report and lead with the two numbers a non-marketer understands: how often the AI names them, and how often it names someone else instead. Visibility Score and Share of Voice do that job.

Then show one worked example. One prompt, the answer text before, the article you published, the answer text after. A single traced example teaches the mechanism better than a page of aggregates, and it makes the next month’s invoice easier.

Keep the sentiment read in. "You are mentioned, and described as the expensive option" is a finding a client will act on, and it is the kind of thing a rank number can never surface. The feature detail covers what each report block contains.

How do you run 20 clients without it eating the team?

Three things do the damage: manual prompt entry, manual publishing, and manual reporting.

Prompt sets are the one-time cost. Build a category template once, then swap the brand and the geography per client. Publishing runs through the client’s CMS connection or a webhook, so week 3 is a review step rather than a copy-paste shift. Reports generate off the same sweep that produced the data, so nobody rebuilds a deck.

That leaves the strategist judgement in week 2, which is the work you should be selling anyway. How the loop runs end to end walks through the mechanics.

Where does measurement stop and content work start?

This is the honest boundary, and it is worth telling clients up front. Measurement tells you which prompts you lose. It does not win them.

The peer-reviewed GEO paper presented at KDD 2024 tested optimization methods on a 10,000-query benchmark and then validated them against Perplexity as a live engine. Adding quotations improved the position-adjusted word count metric by 22% over baseline, while citing sources and adding statistics improved the two metrics by up to 9% and 37%. Keyword stuffing performed 10% worse than doing nothing.

So the content work is specific: fact-dense, source-attributed, answer-first passages. Not volume, and definitely not density.

Beyond the page, evidence gets thinner and you should say so. Ahrefs’ December 2025 study of 75,000 brands found branded web mentions correlated with AI visibility at roughly 0.66 to 0.71, while Domain Rating managed only about 0.27 to 0.33. That is correlational, vendor-produced, and confounded by brand size, so treat it as a lean toward earned brand presence rather than a lever you can invoice against. An agency that says this out loud sounds more credible than one that does not.

What you get on the Agency plan

• Five engines tracked per client, with Google organic rank from the same query
• Visibility Score, Share of Voice, citation gap diagnostics and sentiment
• Article generation that publishes into the client’s own stack and then re-measures the effect
• Site audits and white-label reports
• 10,000 queries, 1,000 prompts, 50 articles a month

Start on the free plan and put one client through a full loop before you sign anything. When the second client makes sense, the plans are here. If you want to see live multi-brand data before you create an account, the public AI visibility leaderboard runs on the same measurement.