GEO vs AEO vs SEO: What Agencies Need to Know Before Pitching AI Search Services

What each stage rewards

Expert tip: stage one is the cheapest place to win and the most commonly broken. Before selling a client any content programme, pull server logs and confirm which AI user agents fetched pages in the past 30 days. A CDN bot rule added by a developer two years ago has silenced more brands in AI answers than any content shortfall.

Google’s official position, and why you should quote it in pitches

Google publishes guidance on optimizing for its AI features, and the guidance is unusually blunt. The short version: optimizing for AI experiences in Google Search is still SEO, and the newer acronyms mostly relabel existing work.

Three specifics from Google’s own documentation are worth memorising, because they settle arguments in sales calls.

1. There is no AI-specific opt-out file

Appearance in AI Overviews and AI Mode is governed by the standard preview and indexing controls that SEOs already know: nosnippet, data-nosnippet, max-snippet, and noindex (Google Search Central: AI features and your website). Those controls are separate from the third-party AI crawler directives such as GPTBot or ClaudeBot.

2. llms.txt is not a Google lever

Google’s AI optimization guide states that site owners do not need llms.txt or similar AI text files for Google Search, including its generative features, and that publishing such a file will not help or harm visibility because Google Search ignores it. Google’s John Mueller has separately described the discovery use case as a dead end. Independent large-scale studies have found no evidence that major AI search systems fetch the file at meaningful rates.

Practical position for agencies: keep llms.txt as a cheap, optional courtesy for non-Google AI services if a client wants one. Never bill it as a ranking or citation lever. Never build a pitch around it.

3. Crawling documentation moved

Google’s crawling and robots reference now lives at developers.google.com/crawling, with IP range files under /crawling/ipranges/. The old googlebot.json file is now common-crawlers.json. Update any internal playbooks and client documentation that still point at the old paths.

Why quoting Google helps you sell: prospects have been pitched by three vendors promising proprietary AI-search secrets. Citing primary documentation, then explaining the small set of things that genuinely differ, positions your agency as the adult in the room. Trust closes deals that hype cannot.

What the citation data shows in 2026

Use numbers in pitches, and attribute every one of them. Below is the evidence agencies rely on most, with sources named so a prospect can verify each claim.

Scale of AI search

How citations get awarded

The four conclusions that should shape every AI search proposal

1. Ranking still matters, but position 1 is not the ticket. Nearly half of AI Overview citations come from pages ranking 5–10, so mid-page rankings become monetisable in a way they were not before.
2. Google is two engines, not one. AI Overviews and AI Mode share a user experience after the I/O 2026 merge, yet they cite different URLs. Score and report on both.
3. Freshness is the most under-sold retainer line item. A scheduled refresh programme has a stronger evidence base than most link campaigns for AI visibility.
4. Off-site mentions do heavy lifting. If a brand is absent from YouTube, Reddit, and Wikidata, on-page work alone will underperform. Plan for digital PR and link outreach alongside content.

The 5-layer AI Search Readiness model

Frameworks sell. More importantly, a framework keeps a delivery team from skipping the boring layers. Here is the model we use on white-label AI search engagements, ordered by leverage. Work top to bottom; skipping ahead wastes budget.

Layer 1 – Access: can machines read the page?

Goal: every system you want to appear in can fetch and parse the content.

• Audit robots.txt for AI user agents and decide each one deliberately.
• Check the CDN, WAF, and bot-management rules. Cloudflare and other providers began blocking AI crawlers by default in 2025, and many site owners never noticed.
• Verify with server logs, not assumptions. Confirm 200 responses for GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot.
• Test rendering without JavaScript. If the primary content disappears, fix rendering before writing a word – rendering and crawl remediation sit inside a standard technical SEO audit.
• Confirm no accidental nosnippet, max-snippet, or noindex directives on pages the client wants surfaced.

AI crawler reference table

Two points clients regularly get wrong. First, Google-Extended controls Gemini and Vertex grounding, not appearance in AI Overviews. Second, user-triggered fetchers ignore robots.txt by design, so blocking them requires server-side controls rather than a text file.

Emerging standard to watch: Web Bot Auth (built on RFC 9421) lets bots authenticate with a Signature-Agent header and a published key directory. Google-Agent already uses the mechanism. Reverse-DNS verification remains the fallback for now.

Layer 2 – Structure: can a passage be lifted cleanly?

Goal: each section stands alone as an answer. Specify the rules at brief stage, not at edit stage – our SEO content briefs carry passage requirements into every draft.

1. Give every H2 and H3 a question or a clear noun phrase, matching how people actually ask.
2. Answer in the first 40–60 words beneath the heading, then expand.
3. Keep citable blocks to roughly 130–170 words.
4. Front-load the most valuable answers. Nearly half of citations come from the first third of a page.
5. Use tables for comparisons, ordered lists for processes, and short paragraphs of two to four sentences.
6. Name entities explicitly. Write “the citation audit,” not “it.”
7. Maintain a strict heading hierarchy – no jumping from H2 to H4 for visual effect.

Layer 3 – Substance: is there anything worth citing?

Goal: the page contains statements a model would rather quote than paraphrase.

• Include specific figures with dates and named sources.
• Link to primary sources: official documentation, standards bodies such as Schema.org and the W3C, government and academic publications, and original research.
• Publish original data. A survey of 200 clients, a benchmark study, or an anonymised dataset of results creates citations no competitor can copy.
• Add first-hand experience: process detail, screenshots, real numbers, and mistakes made along the way.
• Answer the awkward questions competitors dodge, including pricing ranges and limitations.

Layer 4 – Signals: does the web treat the brand as an entity?

Goal: models recognise the brand, not just the domain.

• Complete Organization and Person schema, with sameAs links to every official profile. Validate after every release, or hand implementation to a schema markup service.
• Establish and maintain a presence on Wikidata, LinkedIn, YouTube, and relevant industry directories.
• Earn mentions in third-party roundups and comparison articles where buyers research.
• Participate honestly in communities. Reddit is a heavy citation source across several platforms, and manufactured astroturfing is both against platform rules and easy to detect.
• Keep NAP data consistent everywhere for local clients – a job that pairs naturally with local citation management.
• Add named authors with real credentials and bios.

Layer 5 – Surfaces: are you present where answers get assembled?

Goal: platform-specific presence, measured separately.

• Score AI Overviews and AI Mode separately, because they cite different URLs.
• Maintain Bing indexation and use IndexNow, because Microsoft Copilot leans on the Bing index.
• Ask advocates to add the brand to Google’s Preferred Sources, which is available in all languages as of April 2026, with over 345,000 sources already selected. Google has signalled intent to develop the feature further as a signal.
• Pursue Highly Cited recognition through original primary reporting that other publishers reference.
• Feed Community Perspectives with genuine firsthand content from named people.
• Prepare for agentic tasks: real form elements, semantic labels, stable layouts, and no critical action hidden behind hover-only interactions.

Platform-by-platform playbook

Each engine draws from a different pool. Optimising for one does not deliver the others – only 11% of domains appear in both ChatGPT and Google AI Overviews for the same query.

Expert tip: run the same five prompts across all six platforms once a month and record which domains they appear on. Do it before the pitch, and you walk into the room with a competitive gap analysis nobody else brought.

How to package and price AI search services

Most agencies get pricing wrong in one of two directions. They either bundle AI work into an existing retainer for free – which trains clients to see it as worthless – or they invoice a separate “GEO retainer” that duplicates work already being paid for.

The workable model: one integrated search retainer, with an AI-readiness project at the front and AI-visibility reporting throughout.

A three-tier structure that holds up in sales conversations

Set actual figures against your own market and cost base, and benchmark them against published SEO packages before quoting. The ICP most agencies serve here already budgets $1,500–$10,000 per month for outsourced search work, so the tiers above map onto existing spending patterns rather than asking for a new budget.

Where a white-label partner fits

Building AI search capability in-house means hiring for technical SEO, content strategy, data analysis, and digital PR all at once. Most 5–50-person agencies cannot justify four hires against an unproven line of business.

A fulfilment partner changes the maths. Your team handles strategy, positioning, and client relationships. The partner delivers audits, content, technical implementation, and reporting under your brand. Shrushti’s SEO reseller service and white-label reporting dashboards exist for exactly that split, and a dedicated SEO project manager keeps delivery predictable while your team sells.

Scope-of-work language you can reuse

Included: technical accessibility for named AI crawlers; server-side rendering verification; structured data implementation and validation; passage-level content optimization on agreed pages; entity and profile consistency work; monthly prompt-level visibility measurement across [named platforms]; monthly reporting with commentary.

Not included and not promised: guaranteed placement in AI Overviews, AI Mode, ChatGPT, Perplexity, Claude, Gemini, or Copilot answers; control over how any AI system phrases or attributes an answer; fixed timelines for citation acquisition. AI answer surfaces change without notice, and no provider controls their output.

That second paragraph wins more deals than it loses. A guarantee has already burned sophisticated buyers.

How to pitch AI search without overpromising

The five-step pitch that converts

1. Open with the buyer’s own data, not a trend slide. Show their GA4 referral sessions from chatgpt.com, perplexity.ai, and gemini.google.com. Show the year-over-year change in clicks against impressions in Search Console. Real numbers beat any statistic about the industry.
2. Show the visibility gap. Run 10 buyer-intent prompts across the main platforms—list which competitors appear and which do not. A prospect who reads a competitor’s name in an AI answer about their own category needs no further motivation.
3. Explain the pipeline. Walk through crawl → index → fan-out → retrieve → synthesise → cite. Most buyers have never had the pipeline explained, and the explanation is where authority gets established.
4. Name the 80/20 honestly. “Most of the work is SEO done properly. Some of it is new. Here is the new part, and here is what it costs.”
5. Close on a diagnostic, not a retainer. A fixed-scope readiness audit is an easy yes. The audit findings then justify the retainer with the client’s own evidence.

Discovery questions worth asking

• Which prompts would a buyer type before choosing a provider like you?
• Has your traffic mix shifted – impressions steady, clicks down?
• Do you know whether AI crawlers can currently reach your pages?
• Where does your brand appear outside your own website?
• Who is your named subject-matter expert, and does the site credit them?
• What would a win look like in 90 days, in your words?

Objection handling

Red flags that should stop a sale – yours or a competitor’s

• Guaranteed AI citations or guaranteed AI Overview placement
• llms.txt positioned as the core deliverable
• “Proprietary GEO algorithm” claims with no methodology
• Mention-farming or paid Reddit astroturfing
• AI-rephrased content presented as optimization
• Reporting that shows only prompt screenshots with no baseline or trend

Measurement: what to track, what to promise

Measurement is where AI search retainers live or die. The tooling is immature, the platforms hide data, and clients still expect a number in a dashboard. Handle expectations at the proposal stage.

What the platforms actually give you

A KPI set clients can live with