Your SEO agency wants a separate GEO retainer. What is actually different work, engine by engine?

What does Google say about GEO and SEO?

Google says that its generative AI features in Search run on the same systems as its rankings, so SEO work carries over. Its guide to generative AI features states that "our generative AI features on Google Search are rooted in our core Search ranking and quality systems" (Google's wording).² The same guide says that "the best practices for SEO continue to be relevant" (its words).² On markup, it says that "structured data isn't required for generative AI search, and there's no special schema.org markup you need to add" (its words).²

Google's page for site owners goes further. It states that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" (its words).¹ It adds that "you don't need to create new machine readable files, AI text files, or markup to appear in these features" (its words).¹ So for two of the five engines that matter, a separate GEO retainer has to justify itself against Google's own words.

GEO (generative engine optimisation)

Work that makes a brand more likely to be named, described accurately and cited in answers written by AI engines.

AEO (answer engine optimisation)

The same aim applied to answer features inside a search engine, such as Google AI Overviews and AI Mode. Google treats both as SEO.

The question for a marketing director is narrower than "is GEO real". It is which line items in the proposal describe work that an SEO retainer does not already cover. That differs by engine, so this post takes the engines one at a time.

Why ChatGPT, Copilot and Gemini are a different problem

The three engines outside Google Search retrieve differently, and two of them often do not retrieve at all. Ahrefs, which sells SEO and AI visibility tools, studied 66.6 million free, logged-out ChatGPT responses across all countries in August and September 2026.³ It found that 14.4% of them triggered a web search, with about 1.6 fan-out queries per searching response.³ The rest were answered from the model's memory.

Links are rarer still. In Similarweb's US data covering June 2025 to May 2026, 6.8% of ChatGPT answers carried a citation.⁴ When ChatGPT does search, the prompts have become more search-like. Semrush, which sells AI visibility tools, found in its US clickstream panel that search-enabled ChatGPT prompts averaged 8.7 words in January–February 2026, up from 4.7 words a year earlier.⁵ Non-search prompts shrank from 24.9 to 13.5 words over the same comparison.⁵

Our view: this is the strongest argument for engine-specific work, and it has nothing to do with markup. An answer written from memory cannot be fixed with a page edit this week. It can be moved by what the model has absorbed from third-party sources over time, and by being in the index for the minority of answers that search.

Technical access: which crawler rules are new work?

The new work is a per-bot access policy; the crawlability underneath it is SEO you already pay for. Google says that Googlebot is the control for AI features in Search, because AI is built into how Search functions.¹ If your pages are indexed and eligible for a snippet, AI Overviews and AI Mode can already use them. No new file or tag is involved.

OpenAI runs separate bots with separate rules. Its documentation says OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features" (OpenAI's words), while GPTBot crawls content that may be used in training.⁶ The same page says "each setting is independent of the others" (its words), so a site can allow one and block the other.⁶ Your SEO agency has probably never had to write that policy, and a wrong line removes you from ChatGPT search outright.

The CDN adds a third layer. Cloudflare's July 2026 changelog splits AI traffic into Search, Agent and Training, and from 15 September 2026 new domains block Training and Agent traffic by default on pages that display ads.⁷ Our decision table for blocking training crawlers sets out which bots to keep. The ledger entry here: one policy document and a quarterly check are new; crawl budget, sitemaps and speed are not.

Entity facts and third-party sources: the same work, weighted differently

Off-site work is familiar SEO, but each engine weights the sources differently, and that changes where the hours go. Profound, which sells AI visibility software, analysed 11.84 billion citations across eight engines between 16 April and 16 July 2026.⁸ Company-run sites took 69% of Gemini's citations and 47% of ChatGPT's, while earned media took 30% of ChatGPT's citations against 17.0% for Google AI Overviews.⁸

Breadth differs as much as mix. Semrush's 2026 AI Visibility Index, built on 126 million US prompts from January to April 2026, found ChatGPT cited about 15 sources per response against about 3 for Gemini.⁹ Google's Gemini API documentation describes the mechanism: the model "analyzes the prompt and determines if a Google Search can improve the answer," then runs its own queries.¹⁰ Gemini reads few pages and leans on the brand's own; ChatGPT reads widely and mixes in press and reviews.

For factual questions the engines turn to the official source. In Ahrefs' separate test of 20,000 US prompts, 84.4% of factual questions about pricing, features or company details cited at least one official source.³ So the entity-facts line is the same page work as SEO, with one addition: an engine-by-engine register of which third-party pages each engine actually reads.

Do "best X" and comparison pages need separate GEO work?

They need the same pages written for sub-questions, which is a content brief, not a new retainer. Google says both AI Overviews and AI Mode may use a "query fan-out" technique, issuing related searches across subtopics.¹ A comparison page gets cited when it answers one of those sub-questions well, whether or not it ranks for the headline query.

The ranking overlap is partial. Ahrefs studied 863,000 search results and 4 million cited URLs, published in March 2026 with the data period and geography not stated.¹² In that set, 38% of pages cited in AI Overviews ranked in the top 10 for the same query, data period not stated.¹² Our note on ranking and AI Overview citations explains why that figure is about sub-questions rather than the end of rankings.

Prompt length points the same way. The Semrush finding that search-enabled ChatGPT prompts grew to 8.7 words in early 2026 means buyers now type the whole comparison into the box, with the constraint attached.⁵ Your "best accounting software for a ten-person agency" page has to say so in its opening lines. That is a brief your SEO content team can write. Our view: if the GEO proposal prices comparison pages as a new deliverable, ask what the SEO content line has been producing.

Product feeds: the new line for anyone who sells products

For retailers, the feed is the one on-site deliverable that is genuinely new, because AI shopping answers now retrieve from feeds rather than pages. Profound, which sells AI visibility tracking, analysed shopping prompts tracked by 687 of its customers between 1 July and 24 August 2026.¹³ In its daily data, the share of ChatGPT Shopping picks retrieved through integrated product feeds rose from 8.26% to 61.54% on 10 July.¹³

Google says the same in its own way. Its guide states that Merchant Center feeds and Google Business Profiles "can help your products and services to be visible in both AI responses and other Google Search results" (Google's words).² A retailer with a well-kept Merchant Center feed has done the Google half already; the ChatGPT half is a new integration against OpenAI's feed specification.

B2B companies have an equivalent in their entity feeds: the business profile, the software directories and the review sites engines read. Those are listings work, which most SEO retainers already touch. The feed line is new for product sellers and relabelled for everyone else. Our AI Visibility service treats feed integration as a separate deliverable for exactly that reason.

What is genuinely new in measurement?

Measurement is the line with the least overlap, because no ranking tool reports what an AI engine said. Search Console and rank trackers count positions and clicks. An AI engine can name a competitor, misstate your price or describe you accurately, and none of that appears in a ranking report.

The free sources are partial. Bing Webmaster Tools added an AI Performance report in public preview in February 2026, covering Microsoft Copilot, AI-generated summaries in Bing and select partner integrations, with grounding queries shown as a sample.¹¹ It does not say ChatGPT is covered, and it has no equivalent for Gemini. Google's AI reports in Search Console count impressions in its own features and nothing beyond them.

So the new work is prompt-based: a named list of buyer prompts, run on each engine, repeated because answers vary from run to run, with a margin of error on every share. Ours runs each prompt on ChatGPT, Gemini, Microsoft Copilot, Google AI Overviews and Google AI Mode on three separate days. A single run per engine is a screenshot, not a measurement, and a proposal that offers one should say so.

Our view: this line deserves its own price, and it is the line most often missing from a "GEO retainer" that is really an SEO retainer with a dashboard. Ask who owns the prompt list and whether the report shows accuracy as well as mentions.

How do you test a retainer line by line?

Put every line in the proposal through the same five questions and sort it into new, relabelled or missing. The test takes an hour with the SEO contract open beside the GEO one. Our monthly deliverables checklist covers what should arrive each month; this test covers whether you are paying for it twice.

The sorting matters more than the total. A line marked relabelled is not worthless, it is already bought, so it moves back into the SEO scope at no extra cost. A line marked new gets its own price and its own evidence of delivery. A line marked missing, usually measurement or a feed, is the one to add before signing anything.

Run the test on both agencies if the GEO proposal comes from a different firm. The same relabelling happens in reverse: an SEO retainer that now lists "AI answer optimisation" has often added a word, not a deliverable. The question is always what changes in what an engine reads, retrieves or reports.

What the published-price dataset shows about separate GEO retainers

Most agencies that publish a GEO price bundle it with SEO or content, which makes the "separate retainer" framing unusual rather than standard. In our dataset of published GEO prices, seen on 5–6 October 2026, 34 of 41 agencies listed a price, alongside 25 tools.¹⁴ Many of those agency pages describe GEO as part of a wider search or content retainer rather than a standalone line.¹⁴

The scale of the two markets differs by a factor that matters here. Across the 13 listed agencies pricing in US dollars, the median monthly entry price was $3,400; across the 16 tools with a monthly US-dollar entry tier, it was $79.¹⁴ A separate GEO retainer at agency rates therefore has to deliver something a tool subscription plus your SEO team cannot. Our comparison of published GEO prices lists every row with the date we saw it.

A disclosure, since we are in that dataset too. Sigzen AI sells AI visibility work at published prices.¹⁵ The AI Visibility Audit is $3,500 (₹1.5L) one-time, and the AEO/GEO sprint is from $8,000 (₹3L) over six weeks.¹⁵ Programmes run from $3,500 (₹1L) a month on a six-month minimum.¹⁵ Our programmes are priced as measurement plus engine-specific fixes, and we tell clients to keep their SEO retainer for the on-page work.

Our view: pay for measurement and engine fixes, never twice for on-page SEO

Pay for the lines an SEO retainer cannot produce, and decline any line that renames page work your SEO team already does. The new lines are repeated measurement across five engines, a per-bot access policy, a product feed where you sell products, and a source register weighted by engine. Everything else in a typical GEO proposal has a home in the SEO scope.

The engine ledger decides the split. For Google AI Overviews and AI Mode, Google's own guidance makes the case that the work is SEO, and the additions are a feed and a sub-question content brief.² For ChatGPT, the work is the search minority, the official-source habit and the third-party sources the model absorbs between updates. For Gemini, it is your own site read a few pages at a time. For Copilot, it is Bing's reporting and whatever its partner integrations turn out to be.

Ask your SEO agency a single question before you sign: which of these lines would you still deliver if the GEO retainer were cancelled? Everything on that list is already yours. Everything left over is the real GEO scope, and it is usually smaller, more specific and easier to measure than the proposal suggests.