Search marketing has developed a strange imbalance. We spend more time discussing whether ChatGPT mentions a business than whether its most important service pages rank, attract qualified visitors and generate enquiries. AI visibility matters, and I expect it to matter more. But the attention paid to it is running ahead of the commercial evidence. For a business with a limited marketing budget, that distinction decides what gets done this month and what stays on the experiment list.
The argument is not that search will stay unchanged. It is that we need three connected disciplines now: SEO to make the business discoverable through search, AEO to make its answers useful and extractable, and GEO to improve how it is represented in generative responses. Treating the newest acronym as a replacement for the previous one is an expensive mistake.
What actually happened
On October 21, SEO researcher Marie Haynes published an analysis of twelve months of Google Analytics 4 data from 100 websites of varying sizes, representing more than 145 million sessions. In this sample, visits attributed to ChatGPT equalled 0.62% of Google organic sessions. Add the referrals from other large language model tools and the total increased to 0.73%. These percentages are ratios against Google organic traffic, not percentages of total website sessions. If a site received 100,000 Google organic sessions, the observed proportions would imply roughly 620 ChatGPT sessions and 730 sessions from all measured LLMs together. That illustration is arithmetic, not a forecast for any particular business.
The conversion figures are more interesting, and easier to misuse. Only 55 of the 100 websites were tracking conversions. Within that subset, ChatGPT visitors converted at 13.8%, compared with 3.02% for Google organic visitors. That is more than four times the conversion rate. Yet Google organic still delivered 27 times more conversions than all LLM referrals combined. A high percentage from a small stream does not automatically outweigh a lower percentage from a much larger one.
This is an observational dataset, not a universal market-share study. The public analysis does not provide a detailed industry breakdown, consistent conversion definitions or sufficient methodology to generalize the exact ratios to every sector. It also cannot tell us whether the same visitors would have arrived through Google without their earlier AI interaction. The scale of the observed gap still deserves attention. We can take the direction seriously without pretending 0.62% is a law of marketing.
Why a small team cares
Consider a small company that installs and maintains air-conditioning systems in one region. Its growth depends on a predictable stream of local enquiries: urgent repairs, new installation quotes and seasonal servicing. The owner hears that customers now ask ChatGPT which providers to trust and considers moving most of the search budget into a new GEO retainer. The agency presents screenshots showing the company named in several AI answers. Those screenshots might indicate progress, but they do not establish how many extra booked jobs followed.
SEO remains the baseline because buyers continue to search with commercial intent. A person searching for an emergency repair is looking for an available provider, not a long explanation of the future of search. A person comparing installation options may read several results and return later with a shortlist. Google organic also includes branded searches, local discovery and research-driven visits. The work spans technical accessibility, content quality, on-page structure and authority. None of that becomes obsolete because an assistant can summarize websites.
AEO, or answer engine optimization, asks a narrower question: can a search system or assistant find a precise, trustworthy answer to the customer's question? For the installer, that might mean an understandable section on whether a service visit includes diagnostics, an explanation of maintenance intervals, or a clear policy on weekend callouts. Useful headings, factual answers, visible authorship or business credentials and appropriate structured data can help machines interpret the content. Structured data describes what is actually on the page; it does not guarantee selection into an AI answer or a search feature.
GEO, or generative engine optimization, concerns how the company is discovered, understood and represented in generative responses. A model may assemble a recommendation from business listings, third-party reviews, directory entries, the company's website and other accessible sources. Consistency of its name, services, location and claims therefore matters. So does having information worth citing. There is overlap with traditional SEO and good public relations. The useful GEO activities are not necessarily new tasks; often they are existing quality and credibility work evaluated through a different interface.
That overlap is why I would avoid three entirely separate operating plans. Publishing accurate service information, building credible external references and making pages technically accessible can support all three goals. A small team should maintain one search-content system and add answer coverage and AI representation checks to it. Separate reporting views can be useful, but separate departments, parallel content calendars and three simultaneous optimization packages can multiply costs without creating a better customer experience.
There is a practical distinction between being visible and being valuable. Suppose an agency reports that the installer appears in 30% of selected chatbot prompts, up from 15%. That sounds like a doubling in visibility. But the result could change when prompts are rephrased, when location changes, when the tool chooses different sources or when a model update lands. It is not the same measurement as an enquiry, a qualified sales opportunity or a signed job. Without stable testing conditions and some connection to customer outcomes, the metric is a diagnostic signal rather than a business KPI.
Conversion rates also need interpretation. AI referrals may have a different intent mix from Google visitors. Someone who has already compared suppliers inside an assistant might arrive at a website ready to enquire. Google organic includes early researchers, existing customers and people seeking general information. Those are different journeys. A comparison between 13.8% and 3.02% does not prove that an AI visit causes a higher purchase probability or that buying more AI visibility will reproduce the same rate.
The denominator matters especially for smaller businesses. Five enquiries from 35 assistant referrals can produce an impressive conversion percentage. It can also fluctuate dramatically the following month. Meanwhile, a stable stream of hundreds of search-driven enquiries may support staffing, advertising decisions and revenue forecasts. A good marketing report should show both rates and absolute counts, ideally with qualified leads, resulting sales and value where available.
Measurement has to cover the whole funnel. GA4 can tell us which sessions carried recognizable referrals, provided key events are configured correctly. Search Console can show search queries, impressions and clicks. A CRM or even a disciplined lead register can connect enquiries to booked jobs. Asking a new customer how they found the business can uncover AI-assisted research that analytics missed, although self-reported answers are imperfect too. Together, these sources offer a more useful account than a standalone count of AI mentions.
For the air-conditioning company, I would compare monthly Google organic enquiries, local business-profile leads, measurable assistant referrals and genuinely qualified sales. I would also inspect whether the main commercial pages are converting. A service page gaining traffic but losing enquiries might need better pricing context or a simpler quote form. Fixing that could outperform weeks spent pursuing citations in AI-generated answers. The right priority is whichever change improves the commercial system, not whichever dashboard looks newest.
The allocation of effort should reflect a company's current starting point. A business with inaccessible pages, thin service descriptions and no reliable conversion tracking has little reason to purchase an expensive standalone GEO programme. A business with strong SEO, clear attribution and evidence that buyers routinely use assistants to compare providers has a better case for a focused generative-search experiment. The goal is not a fixed industry-wide percentage of budget. It is protecting proven demand capture while reserving enough capacity to learn about the next interface.
Hype vs useful
The weakest promise in AI-search marketing is that visibility automatically turns into traffic and revenue. Some assistants cite sources and provide links. Others give a complete response, leaving the user no reason to click. An appearance in an answer may improve familiarity, but its financial contribution is often unobservable. Vendors can sell tracking for brand mentions, share of voice and citation frequency. Those tools can be useful for diagnostics and competitor research. They should not be reported as equivalent to sales, and the cost of collecting the metrics must make sense for the business.
The opposite claim is also too simple: because measured referral traffic is tiny, GEO can be ignored. Discovery can influence purchases without appearing as a referral, and a changing distribution channel can become important before it is easy to attribute. Business information that is clear, current and corroborated across the web gives customers and machines fewer opportunities to misunderstand an offer. There is little downside to improving that information when it also helps human buyers.
I would be particularly skeptical of large volumes of so-called AI-optimized content designed only to repeat the same answers in slightly different wording. If ten pages offer no distinct expertise, pricing clarity, local relevance or firsthand evidence, they do not become valuable merely because they were written for a language model. Such programmes create ongoing editorial and technical maintenance while adding noise to the website. A focused set of useful pages is easier to keep accurate and easier to connect to business outcomes.
There are sensible tests. Monitor a repeatable sample of commercial questions customers actually ask. Record whether the brand is represented accurately, which sources are cited when citations exist, and whether competitors appear for relevant reasons. Improve missing facts on the website and trusted third-party profiles. Then observe changes over several weeks alongside enquiries and assisted-sales evidence. A change in chatbot outputs without a corresponding customer signal is still learning, but not yet proof of return.
The test must also account for the fact that AI output is variable. One answer on one afternoon is not a market benchmark. Prompts, geography, model choice, conversation context and the tool's live retrieval can all change results. A credible report specifies its questions and conditions and avoids presenting temporary answer placements as guaranteed rankings. If the testing process becomes more expensive than the potential benefit, reduce its scope.
The most defensible investment is shared infrastructure: fast accessible pages, accurate service data, original demonstrations of expertise, recognizable business identities, useful comparisons, authoritative references and working conversion tracking. These strengthen search performance now and make information easier to reuse in AI experiences. When the same asset helps SEO, AEO and GEO, the business pays once for something with multiple paths to value.
Agency pricing should follow this logic as well. I would rather sell an outcome-oriented search programme with a modest, clearly scoped AI-discovery component than three disconnected retainers named after acronyms. Clients need to know which pages generated leads, what was improved, where their brand information was inconsistent and what the next experiment costs. The work becomes easier to defend when the monthly report distinguishes confirmed business results from early visibility indicators.
There is also a strategic reason not to abandon SEO: the platforms are connected. AI search systems may retrieve or learn from discoverable web content, while Google increasingly delivers AI-generated experiences inside its own search products. That does not mean traditional rankings automatically determine generative citations. It means an accessible, trustworthy web presence is still a useful foundation even as interfaces change. The exact weighting of sources, and the value of appearing in an answer, remains platform-dependent and partly opaque.
For now, the commercial verdict is straightforward. Do not move the bulk of a working search budget to GEO because an assistant occasionally mentions a competitor. Keep improving the pages and listings that support real demand, measure what turns into qualified business, and add targeted AEO and GEO work where it improves clarity, credibility or measurable opportunities. AI discovery deserves a place in the plan. It has not earned the right to replace the plan.
