How to Sell AEO and GEO Services to Your Existing SEO Clients

The way to sell AEO and GEO services to existing SEO clients is to sell an additional, measurable scope of work, built on a visibility gap you can show them. Run a diagnostic first, put the evidence in front of the client, then propose work that answers that gap. Renaming the SEO retainer "AI search" without changing the deliverables is not an offer; it is a price rise.

This guide is for agency owners and account managers who want a service they can defend on a call with a finance director. It covers which clients are ready, the diagnostic to run before you pitch, a worked example, three service packages, a 90-day delivery plan, a pricing method, the sales conversation and the reporting that keeps the client. When you are ready to write the document itself, the companion guide How to Build an AEO/GEO Proposal Clients Can Understand and Approve has a complete, copyable proposal.

What do AEO and GEO actually mean?

The terms are newer than the work, and the industry does not use them consistently. Treat the definitions below as working definitions you can explain to a client, not settled standards.

Answer engine optimisation (AEO) is the work of making a business's information easy for answer-style results to find, extract and repeat accurately. It started with featured snippets and voice answers, and now usually covers AI answers too.

Generative engine optimisation (GEO) is the work of improving how often, and how accurately, AI assistants such as ChatGPT, Gemini and Google's AI Mode mention a brand or cite its pages when they generate an answer. The term was formalised in a 2023 research paper, GEO: Generative Engine Optimization, which also proposed measuring visibility inside generated answers.

In practice, most agencies use the two words for the same job: find out what AI answers say about a client's category, then improve the evidence those answers draw on. The overlap with SEO is large. Google says its AI features use the same foundations as Search: a page must be indexed and eligible for a snippet, and no additional technical requirements or special optimisation are needed to be shown as a supporting link in AI Overviews or AI Mode. OpenAI's ChatGPT search relies on its own crawler, and sites that block OAI-SearchBot are not shown in ChatGPT search answers.

So what is actually new? Three things:

• A new measurement layer. Rankings don't tell you whether an assistant names the client. You need sampled answers, mention and citation rates, and the competitors and sources that appear instead.
• A wider source map. Assistants often cite review sites, directories, comparison articles and forums. Some of the work happens off the client's site.
• Answer-shaped content. Pages that state facts plainly (who it is for, what it costs, where it operates, how it compares) are easier to quote correctly.

That is the honest pitch: the same foundations, plus measurement and work that a standard SEO retainer usually doesn't include.

Which clients should you offer it to?

Not every SEO client needs a separate AEO/GEO service. Sorting your book before you pitch protects your credibility, because the first client who buys a service they didn't need will tell you so at renewal.

A useful test: can you name ten questions a buyer would ask an assistant before choosing this client, and would the client care which competitors get named? If you can't, don't pitch yet.

How do you run a diagnostic before you pitch?

The diagnostic is the asset that makes the sale. It turns "AI search is changing everything" into "here is what ChatGPT, Gemini and AI Mode said when we asked the questions your buyers ask". Build it in six steps.

1. Start with the business goal. More qualified demos, more booked jobs in two new suburbs, fewer lost deals to one competitor. Write it down; it decides which questions matter.
2. Write the buyer questions. Aim for 20 to 40 questions across four intents: category ("best invoicing software for plumbers"), problem ("how do I stop chasing late payments"), comparison ("Tallyworks vs Ledgerline") and fit ("invoicing app that works offline"). Use sales call notes, search queries from Search Console and the client's support inbox. Don't put the client's name in category or problem questions; that tests recall, not discovery.
3. Choose engines and markets. Pick the assistants the client's buyers plausibly use, and the country or language. A UK accounting firm and a US software company need different settings.
4. Agree the competitor set. Three to five named competitors the client actually loses deals to, not everyone in the category.
5. Collect answers repeatedly. Ask each question several times on each engine, on more than one day. Record whether the client is named, whether a client page is cited, which competitors are named and which sources are cited.
6. List cited pages and gaps. Which URLs do assistants cite for each question group? Which of those mention competitors but not the client? Which client pages are cited, and are the facts on them correct?

Why repeated sampling matters

AI answers vary from one ask to the next. In a January 2026 study, SparkToro and Gumshoe found a less than 1 in 100 chance that ChatGPT or Google's AI would return the same list of brands in two responses to the same prompt, and the order was even less stable. The study's author concluded that a visibility percentage across many prompts, run multiple times, is a reasonable measure, while a single "ranking position" in AI is not.

So report rates from samples, with a range, and never a screenshot of one answer. With three answers per engine per question each week, 24 questions on three engines produce 216 answers a week. That is enough to see large differences between brands and question groups. It is not enough to prove that a two-point change is real.

Be clear with the client about what a sample is: the answers your tool collected under fixed settings. It is not a census of what every user saw, and it doesn't include personalised answers inside logged-in accounts with history and memory.

Serpify's AI Visibility is built for this step: you add the questions, it collects several answers per engine every week from ChatGPT, Gemini and Google AI Mode, and reports mention and citation rates with 95% ranges, the competitors named and the sources cited. You can also run the diagnostic by hand with a spreadsheet; it just takes longer to repeat.

A worked example: Fieldnote Digital and Tallyworks

Fictional example. Fieldnote Digital, Tallyworks, Ledgerline, InvoiceHub and every figure below are invented to show the method. They are not Serpify customers or real results.

Fieldnote Digital is a six-person SEO agency. One retainer client, Tallyworks, sells invoicing software to trade businesses: plumbers, electricians and builders. Tallyworks ranks on Google's first page for several trade invoicing terms, but its head of sales keeps hearing prospects say "ChatGPT suggested Ledgerline".

Fieldnote's account manager agreed a goal with the client (more demo requests from trade businesses), wrote 24 questions across four intents, named two competitors, and collected answers from three engines, three answers per question per engine, in two weekly runs: 432 answers in total.

Across the same 432 answers, Ledgerline was named in 198 (45.8%) and InvoiceHub in 150 (34.7%).

The cited sources told the more useful story. Category answers mostly cited two software review sites and a "best apps for tradespeople" article that listed Ledgerline and InvoiceHub but not Tallyworks. Tallyworks's own pricing was published only inside an image, and its review-site profile still described a product the company retired last year.

From findings to a scoped recommendation

Here is how Fieldnote turned that into a recommendation without overclaiming:

• What the data shows: Tallyworks is rarely named when buyers ask category and problem questions, and is named most when the buyer already knows the brand. Competitors appear far more often across the same questions.
• What it doesn't show: that the review-site profile or the missing pricing text caused the gap. They are plausible contributors, chosen because they are visible in the cited sources and cheap to fix.
• What Fieldnote proposed: a fixed implementation project focused on category and problem questions (plain-text pricing, a page for each main trade, a corrected review-site profile and outreach to the article that cites competitors), followed by monthly monitoring of the same 24 questions so both sides could see whether the rates moved.

Notice what was left out: the Google rankings work. Tallyworks already pays for that.

How should you package AEO and GEO services?

Three packages cover most agency situations. Keep them separate so the client can buy the first without committing to the third.

1. The diagnostic (one-off)

• Deliverables: agreed question set, baseline report (mention and citation rates by engine and question group, with ranges), competitor comparison, cited-source map, list of factual errors, prioritised recommendations.
• Effort drivers: number of questions, engines and markets; how many competitors; whether the client's sales team joins a question workshop.
• Boundaries: no changes to the site; recommendations, not implementation.
• Buyer outcome: a clear answer to "are we missing from AI answers, where, and to whom?" before spending more.

2. The implementation project (fixed scope)

• Deliverables: named pages created or rewritten, factual corrections, structured data that matches visible content, review and directory profile fixes, outreach to specific cited sources, crawl access checks for search and AI crawlers.
• Effort drivers: number of pages, how many third-party profiles, client approval speed, legal or compliance review.
• Boundaries: a list of items, each with acceptance criteria. No promise that any assistant will mention the client.
• Buyer outcome: the evidence AI answers draw on is complete and correct, and each change is documented.

3. Monitoring and ongoing improvement (monthly)

• Deliverables: the same question set sampled every week, a monthly report, a fixed number of improvement hours, a review call, and an updated priority list.
• Effort drivers: question set size, number of engines, how many hours of improvement work, reporting format.
• Boundaries: hours that don't roll over unless you say so; new pages beyond the allowance are quoted separately.
• Buyer outcome: the client sees whether visibility changes over time, and the work keeps following the evidence.

A note on what to call each package. Lead with the outcome ("AI visibility baseline", "answer evidence project") and put the acronym second. Most buyers don't care whether you call it AEO or GEO; they care whether competitors are recommended instead of them.

What is new work, and what is already owed under SEO?

This is where agencies lose trust. If your SEO retainer already includes "technical SEO, on-page optimisation and two articles a month", you can't bill the same work again under a new label.

Two rules keep you honest. First, when an AEO/GEO deliverable overlaps with the SEO scope, use the SEO allowance first and say so in the proposal. Second, list what the new scope excludes, in writing.

A 90-day delivery plan

Use a fixed plan with an owner, an output and a review point for each phase. Note what is missing: a target uplift. You can commit to doing the work and measuring it, not to how assistants will answer.

The day-90 report should be equally useful if nothing moved. "We fixed these nine items, the mention rate range overlaps the baseline, and here is what we would test next" is a legitimate result.

How should you price AEO and GEO work?

Price from your costs and the scope, not from what you think the market will bear. The method:

1. Estimate hours per deliverable, by role.
2. Multiply by your fully loaded cost per hour (salary, overheads, non-billable time).
3. Add tooling: the share of your AI visibility tool and other software this client uses.
4. Apply your target gross margin: price = cost ÷ (1 − margin).
5. Keep one-off fees (diagnostic, implementation) separate from the recurring monitoring fee.

Illustrative calculation only. These are not market rates and not Serpify prices. They assume a fully loaded cost of $60 an hour and a 50% target gross margin. Use your own numbers.

Round the result if you like, but keep the working: when a client pushes on price, you can remove scope (fewer questions, fewer pages) instead of cutting your margin.

Two pricing traps. Don't price monitoring so low that it can't include any improvement hours; a report with no work behind it gets cancelled. And don't hide the diagnostic inside the project fee. A paid diagnostic qualifies the client and makes the rest of the proposal easier to approve.

How do you have the sales conversation?

Clients buy when they see their own situation in the evidence. Here is a realistic conversation, using the Tallyworks example, between Fieldnote's account manager (AM) and Tallyworks's marketing director (MD).

The client's goal. AM: "Last quarter you said the target is more demo requests from trade businesses, and that sales keep hearing Ledgerline's name. Is that still the priority?" MD: "Yes, very much."

The problem. AM: "We asked ChatGPT, Gemini and Google AI Mode 24 questions your buyers ask, three times each, twice. When someone already knows your name, you're named in 38% of answers. When they ask for the best invoicing software for trades, it's about 8%, and Ledgerline is named in most of them. That's the gap your sales team is hearing about."

The guide. AM: "We've run your SEO for two years, so we know which pages exist and which facts are missing. The answers cite a few review sites and one comparison article, and your profile on the main review site still describes the old product."

The plan. AM: "Three steps. Fix the evidence: plain-text pricing, a page per trade, the review profile and the article that leaves you out. Measure the same 24 questions every week. Review together at day 90."

The action. AM: "The project is a fixed fee, and monitoring is monthly after that. I'll send a proposal with every item listed, including what your current SEO retainer already covers."

The cost of doing nothing. AM: "If nothing changes, prospects who start in an assistant keep seeing Ledgerline first and arrive at your demo already comparing you to them, or don't arrive at all."

What success looks like. AM: "In three months you'll know, with numbers, whether you're named more often for category questions, whether the facts about you are right, and which work moved them. Then you decide what's worth continuing."

Handling the objections

"We already pay you for SEO." "You do, and this proposal doesn't charge again for anything in that scope. Here's the overlap table. The new work is the answer sampling, the off-site corrections and the pages beyond your current allowance."

"Prove it will pay for itself." "I can't prove that before we start, and I'd be wary of anyone who claims to. What I can do is measure it: we'll report AI referral visits and demo requests from them where your analytics can identify them, alongside mention rates. The diagnostic is small, so you can decide with evidence before committing to more."

"Can you promise we'll be mentioned in ChatGPT?" "No. Nobody controls what an assistant says, and answers change from one ask to the next. We can make the evidence about you complete and correct, and show you whether the mention rate moves."

"Why pay monthly? Can't we do it once?" "You can stop after the project. Monitoring is for seeing whether changes hold as engines and competitors change, and the monthly hours go into whatever the data shows next. If the report stops being useful, cancel it."

"What if the numbers go up and down?" "They will. That's why we report ranges from repeated samples rather than single answers, and why we look at trends over weeks. A small move inside the range is noise, and we'll say so."

What should the monthly report include?

Keep these measures separate. Mixing them is how agencies end up claiming credit they can't support.

For Google's AI features there is no separate channel to point at: Google includes AI Overviews and AI Mode traffic in Search Console's overall Performance report, within the Web search type. For assistants that send referral visits, look for their domains (such as chatgpt.com) in your analytics referral sources, and say clearly that many answers produce no click at all.

Serpify can carry much of this: branded PDF reports and client links for each project, AI Visibility next to Google rank tracking in the same project, and a weekly action plan built by fixed rules. Be aware of its limits too: it reports what the sampled answers said, and it doesn't measure whether a completed task changed the result. The work log and the judgement are still yours.

Honest limits to tell clients

• Answers are samples. Rates describe the questions, engines and settings you tracked, not every user's experience.
• No one controls the output. Not you, not the client, not any tool. Anyone who promises placements in AI answers is selling something they can't deliver.
• Correlation isn't cause. If the mention rate rises after a page goes live, the page may have helped; so might a model update or a competitor's change.
• Engines change. Models, retrieval sources and interfaces change without notice, which can move rates in either direction.
• Some value has no click. A buyer may shortlist the client from an answer and arrive later through branded search. Treat referral data as a floor, not the full picture.

Agency checklist before you pitch

• The client fits: question-led buying, solid SEO basics, competitors visible in answers.
• Goal written down in the client's words.
• 20 to 40 buyer questions across category, problem, comparison and fit.
• Engines, market and three to five competitors agreed.
• At least two weekly runs, several answers per question per engine.
• Rates reported with ranges; no single-answer screenshots as evidence.
• Cited-source map and factual error list.
• Overlap with the SEO retainer listed and not charged twice.
• Separate one-off and monthly fees, built from hours, tooling and margin.
• No promised placements, uplifts or causal claims.

Your next step

Pick one client from the "offer the diagnostic now" row and build their baseline this week. Write 10 buyer questions, name three competitors, and collect answers from ChatGPT, Gemini and Google AI Mode.

The fastest way to do that is to start a Serpify workspace with AI Visibility. New accounts get a 14-day trial of the advanced features, no card required, which covers 10 AI questions, enough for a first look at one client. A full diagnostic with 20 to 40 questions needs a paid plan. Then take the evidence into the AEO/GEO proposal template and turn it into a scope the client can approve.