More agencies are packaging GEO as a separate service with its own line item. We are not going to sell you that. Google’s own documentation is the clearest public confirmation of why, and the rest of this page is the reasoning it does not cover.
Google publishes a guide on optimizing for its generative AI features, updated July 2026. On the question of whether GEO is a separate discipline, it is unusually direct: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” The same page tells you to check any third-party GEO advice against its official guidance before buying it.
That position is Nick Rizkalla’s, and it costs us the GEO retainer we could otherwise package and sell. Below is the evidence, including the parts that cut against us.
Where AI visibility is settled, and where it is not
The pitch you are hearing treats “AI search” as one thing. It is at least three, and they behave differently.
Google’s AI features (AI Overviews, AI Mode). This is the one with published rules. Google says its generative features are rooted in its core Search ranking and quality systems, using retrieval-augmented generation to retrieve relevant, up-to-date web pages from the Search index, plus query fan-out, where the model generates concurrent related queries to fetch additional results. The eligibility rule is concrete, and it has two parts. A page must be indexed and eligible to appear in Google Search with a snippet, and the site must be included in Search generative AI features. That second setting is worth checking rather than assuming, because a site excluded from those features earns no AI visibility no matter how good the content is. For this surface, being crawlable, indexed, and included is not a GEO tactic. It is the entry condition.
Other AI platforms (ChatGPT, Perplexity, and the rest). Different plumbing, and the vendors say so. OpenAI states that ChatGPT will choose to search the web based on what you ask, meaning retrieval is conditional rather than automatic, and that ChatGPT search leverages third-party search providers as well as content provided directly by its partners. What that establishes is narrower than the pitch implies, and worth stating precisely: ChatGPT Search does not run on the publicly documented eligibility and ranking rules Google publishes for AI Overviews and AI Mode. There is no equivalent published rulebook to work against.
Answers with no live retrieval at all. When a model replies from what it absorbed in training, there is no fetch to be included in. Publishing a page does not give that specific response a source to pull from, though it may matter for a later web-enabled answer or a future model.
A service covering all three is not automatically wrong. It owes you an explanation of which mechanism it is addressing on each surface and how each result gets verified. What we optimize is the surface where the rules are published and eligibility is verifiable: Google’s.
What the research actually found, and what it did not
The term GEO comes from a paper by Aggarwal and colleagues, presented at KDD 2024. It is a real piece of work and it is worth reading, but the industry is quoting it in ways the authors did not.
The paper reports that GEO methods can boost visibility by up to 40% in generative engine responses, with citations, quotations, and statistics among the effective methods. “Up to” is doing real work in that sentence. It is a ceiling from a benchmark, not an average you should expect.
Three limits worth carrying with the number:
It is a model, not a map of today’s platforms. The paper presents its two-stage architecture as a representative workflow which, at the time of writing, closely resembles the design of BingChat, and scopes itself to single-turn Generative Engines. A 2023 approximation of one product. Not the blueprint for ChatGPT, Perplexity, or AI Mode in 2026.
The visibility metric is not leads. The authors measure impression metrics like normalized word count and a position-adjusted count that decays by citation position. That is how much of the answer your source occupies. This paper did not measure clicks, calls, booked jobs, or revenue, and it did not test whether anyone hired anybody.
The famous number is one cell in one table. The 115.1% figure that gets quoted as proof that structure beats ranking is the Cite Sources treatment at rank five, in that benchmark. The paper’s own headline is “up to 40%,” and it stresses that the efficacy of these strategies varies across domains. It is an experimental finding, not a forecast for your service page.
What survives is still useful: content that cites sources, quotes real expertise, and states specific facts did measurably better in a controlled test. We apply those findings inside the SEO work rather than packaging them separately, because the study establishes a visibility effect in a benchmark, not a local-business outcome or a way to measure one. That is our packaging decision, not something the paper proves.
What we actually do about it
Google’s guide is refreshingly specific about what does not work, and it matches what we have refused to sell. Google says you can ignore llms.txt files and other special markup, because Google Search does not use them. It says there is no requirement to break your content into tiny pieces for AI. It says structured data is not required for generative AI search. And on the tactic we get asked about most: seeking inauthentic mentions across the web is not as helpful as it might seem.
That last one matters commercially, though note the scope: Google’s guidance covers Google Search, not what enters a future training run. The service marketed as “authority statement PR,” seeding claims about your business across the web so models absorb them, runs together three separate mechanisms: what enters a future training run, what gets retrieved live, and what feeds entity databases. The test we apply is simple. A provider should be able to name which mechanism it expects to move and show how that effect gets observed. If it cannot answer both, we will not sell you the work, because we would be billing you monthly for something neither of us can see.
What we do instead, in priority order:
Make the page eligible. Indexed, crawlable, servable with a snippet. Google’s stated precondition for its AI features, and the one most Nashville sites we audit fail somewhere.
Write what a model cannot get elsewhere. Google draws the line at commodity content, contrasting “7 Tips for First-Time Homebuyers” against “Why We Waived the Inspection and Saved Money: A Look Inside the Sewer Line”. That distinction is the whole local opportunity. You have first-hand knowledge of your trade in this market that a general directory has no way to acquire.
Structure it for a reader. Google’s advice is to write for your human audience with clear headings and sections, not to reformat for machines. Clear hierarchy and direct statements make information easier for retrieval systems and people to locate, though no structure guarantees citation.
Keep the local surfaces current. Google notes that generative responses can include local business information and that Google Business Profile can help your services be visible in both AI responses and other Search results.
Nothing on that list is exotic. That is the argument.
The question to ask any agency selling you GEO
Ask what share of your leads comes from AI, and how they know. The honest answer has four layers, and they are not interchangeable:
- Google AI visibility, separately observable. Google now offers a Generative AI performance report covering AI Overviews and AI Mode. Read the fine print with us: it reports impressions, grouped by page, country, date, and device. Impressions are how many times links to your site were shown. Not clicks. Not calls. Not leads. Google is also explicit that it is rolling this out to a subset of website owners, so your property may not have it at all.
- Google AI traffic and leads. Not derivable from that report. The impression is where the evidence stops.
- Chatbot referrals. Traffic arriving from ChatGPT or Perplexity, partially visible in analytics.
- AI-assisted decisions. Someone reads an AI answer, then searches your name, then calls. Standard analytics cannot reliably connect the two, so the size of this layer cannot be established from your reporting. A CRM source field or an intake question can give you partial self-reported evidence, which is worth collecting and is not the same as measurement.
An agency that quotes you a single confident “AI lead percentage” has collapsed four layers with very different levels of observability into one number. Google’s own warning is worth repeating: be wary of third-party tools that promise ranking success or claim to use internal Google metrics, because no third-party tool has access to Google’s internal ranking or AI systems.
Where our position gets tested
The plumber example is the easy case, and we should say so. Someone with water on the floor at 9pm is not researching. They tap the top map result and call.
The harder case is the one worth thinking about: a cosmetic dentist, an estate attorney, a remodeler, a B2B supplier. Those buyers research for weeks, and some of that research now happens inside an AI tool. If your firm surfaces in that early conversation, you are on the shortlist before the first call.
Our position holds there too, for a different reason. Those questions are the ones a general directory answers generically, and you can answer them from having done the work. Publishing specific, first-hand operational knowledge a directory does not have improves your chances of being a useful source. It improves eligibility. It does not guarantee selection. And it is the work, not a product.
Our position would have to change for a business whose qualified buyers rely materially on AI platforms that Google-centered SEO does not reach or measure well. We have not audited a Nashville business where that was true. If yours is, we will say so, and the honest answer might be that we are not the right shop.
Common questions
Does GEO replace SEO? For Google’s AI features, Google says optimizing for them is still SEO, and that they rest on core Search ranking. For other platforms, retrieval sources and ranking vary, and some answers involve no live retrieval at all. Rank Nashville does not package either as a separate retainer: for Google, the work is SEO; for the other platforms, we will not promise a repeatable method until the mechanism and the measurement are clear enough to defend.
Should a Nashville business invest in GEO right now? Invest in being eligible, specific, and genuinely useful. An agency selling GEO on top of a site with unresolved crawl problems has the order backwards.
Is it late? No, and be skeptical of urgency as an argument. Being more useful and more specific on your own subject is an advantage a small business can actually control, even though it guarantees nothing, and it is as available to a five-person shop in Donelson as to anyone.
How would you measure it for me? With what exists, which is less than you are being told: impressions where the report is available, some chatbot referrals when the referrer survives the trip, and self-reported answers for everything else. If anyone tells you a synthetic-query tool has measured your AI visibility, ask which platform, which model, on what date, how many repetitions, and whether the result was a brand mention or a real citation. Those questions have answers or the measurement is decoration.
Where to start
Request a free audit. On this question we check four things: whether your pages are indexed and eligible to be served with a snippet, whether your site is included in Search generative AI features and whether that report is available to your property at all, whether your pages say anything specific enough to give a model or a person a reason to cite them, and whether your claims survive a check.
If AI visibility is a real gap, we will show you the evidence. If it is not, we will say so and point at what is actually costing you leads, which in most audits we run is something duller and more fixable.
That is the same Nashville SEO work we have always done. Google just published a guide supporting that position for its own AI Search features. The rest of the argument is ours to defend.
Related reading: How AI Overviews Reshape SEO and User Intent, on what happens when the AI answer sits above your listing.
Call Nick Rizkalla at (615) 988-1309 for a free audit.
Rank Nashville, 615 Main Street, Suite 123, Nashville, TN 37206, (615) 988-1309
Sources
Google Search Central. “Optimizing your website for generative AI features on Google Search,” updated July 10, 2026. The GEO-is-still-SEO position, RAG and query fan-out, eligibility requirements, the commodity content distinction, and the mythbusting section on llms.txt, chunking, structured data, and inauthentic mentions.
Google Search Console Help. “Generative AI performance report (Search).” Covers AI Overviews and AI Mode. Reports impressions by page, country, date, and device. Rolling out to a subset of website owners. Does not report clicks, conversions, leads, or revenue.
Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., Deshpande, A. “GEO: Generative Engine Optimization,” KDD 2024, arXiv:2311.09735. The up-to-40% finding, the representative-workflow and single-turn scoping, the impression metrics, and the domain variation.
OpenAI. “Introducing ChatGPT search,” October 31, 2024, updated February 5, 2025. Conditional web search, third-party search providers, partner content.
Not established by anything cited here, including by us: the share of leads a Nashville business currently receives through AI answers. We describe which layers are observable and which are not, rather than quoting a number nobody can support.