SEO, AEO, and GEO: how customers find your business

A prospective customer searches for a service, reads a short answer, and asks an AI assistant to compare the options. Your business could appear at every step. It could also be mentioned without receiving a visit, cited for the wrong claim, or discovered through a page that never ranks first for the original question.

SEO, AEO, and GEO are useful ways to examine that journey. The challenge is deciding which work helps the customer and which evidence shows that it helped your business.

For a professional-services firm, our recommendation is to manage one coherent body of public information, then measure how search engines and assistants use it. Start with what customers need to decide, make the evidence accessible, and track the path from discovery to a qualified inquiry.

SEO, AEO, and GEO compared

Search engine optimization (SEO) improves how search engines discover, understand, and select relevant pages. Answer engine optimization (AEO) focuses on making a specific answer easy to find and understand. Generative engine optimization (GEO) focuses on how information is selected and represented in generated answers.

These are overlapping working definitions. AEO and GEO are not universally standardized categories, and an answer can be both direct and generated. Google explicitly treats optimization for its generative search features as SEO. Google’s current AI optimization guide.

The same page can support all three. A well-documented cost comparison might earn an organic visit, supply a concise answer, and contribute evidence to an assistant’s recommendation.

AEO also includes answer formats that do not require generative synthesis. Google’s featured snippets are one example. Google selects those automatically; a publisher cannot mark a page as the featured snippet. Google’s featured snippet documentation.

Follow one customer through the three experiences

Imagine a finance director researching how to review AI-generated financial models. This is an illustrative journey, not a measured customer case.

They first search for a financial-model review checklist. A useful page names the problem, explains who the checklist is for, and can be discovered through relevant links. That is an SEO concern: can the page reach someone with the right need?

Next, they ask, “Which parts of an AI-generated model should a person check?” A clear opening answer identifies input assumptions, formula consistency, reconciliations, scenario behavior, and the source of each material figure. The rest of the page explains those checks. That is an AEO concern: does the content answer the question without forcing the reader to extract it from vague prose?

Finally, they ask an assistant to compare several approaches to financial analysis. A worked example with a downloadable workbook, documented checks, and explicit limitations gives the assistant concrete information to consider. That is a GEO concern: is the evidence selected, attributed, and represented faithfully?

The strongest version of that page serves the finance director at each stage. It does not need three near-identical versions carrying different acronyms. Our Moolaski financial analyst skill provides context for this kind of workflow; the example here does not claim a measured search or citation improvement.

Being found and being cited are different events

Consider the sequence: a system can access a page, retrieve it for a question, use a passage, cite it, and send a visitor. The visitor may then become a qualified inquiry and eventually a customer. Each transition can fail independently.

A system might retrieve your page but prefer another source. It might cite your page while making a claim the page does not support. A correct citation might satisfy the reader’s question without producing a click. A visitor might have no intention of buying your service.

Research supports examining the whole sequence. The 2026 SAGEO Arena preprint evaluated retrieval, reranking, and answer generation together. In its benchmark, some body-text rewrites reduced retrieval performance, even when optimization aimed to improve generated-answer visibility. Its lexical retrieval setup is not a replica of Google or ChatGPT, so its results should guide questions to test rather than become a production ranking formula. SAGEO Arena paper.

For a firm commissioning this work, ask which stage a proposed change targets and how success will be measured. “More AI visibility” is too broad to tell you whether customers are finding the business, understanding it correctly, or contacting it.

What the famous GEO research result actually means

The original GEO paper reported improvements of roughly 40% in answer-visibility measures under some experimental conditions. That is frequently a more exciting headline than the methods justify for a business audience.

The main setup supplied five retrieved sources to a language model. Its Perplexity experiment provided source text as uploaded files and constrained answers to those sources. The researchers measured representation in generated answers, not sales or qualified leads. They did not establish that publishing those edits on a live website would produce 40% more traffic. Original GEO paper, including Appendix C.1.

The practical lesson is narrower and useful: presentation can influence which supplied information appears in an answer. A business still needs evidence that its pages are discovered, that the description is accurate, and that any additional attention is commercially valuable.

Treat benchmark percentages as findings about the experiment that produced them. Ask a provider to show the underlying measurement before accepting a traffic or revenue forecast.

What to improve on your website first

Start with the questions a prospective customer must answer before making contact: what you do, who it is for, what an engagement involves, where you operate, and what evidence supports your claims.

Then improve the material behind those answers. A comparison becomes more useful when it includes assumptions and conditions. A case study becomes more credible when it explains the baseline, measurement method, exceptions, and ongoing work. A technical guide becomes easier to evaluate when readers can inspect an example rather than rely on adjectives.

For Looski’s domain, that could mean explaining which integrations receive a document, showing how a financial workbook is checked, or listing the operating costs omitted from a hardware-only comparison. These are content recommendations, not claims that a particular format guarantees a citation.

Make the important information available as readable page content. Use headings and tables where they help the reader. Give each coherent topic a clear home and link related explanations. Keep material claims current and make the next step obvious for someone whose needs fit your service.

Google says pages must be indexed, eligible for a snippet, and included in its Search generative AI features to qualify for those experiences. It does not require special AI schema or a prescribed content length. Meeting the requirements does not guarantee selection. Google also says it ignores llms.txt for search visibility and ranking; such a file can have other documentation uses. Google’s guidance.

Decide deliberately what crawlers can access

Public marketing content and confidential client material serve different purposes. Make that distinction before changing crawler settings.

OpenAI documents separate bots for search and potential model training: OAI-SearchBot and GPTBot. Their preferences can be configured independently. A blanket block on every AI crawler can therefore have different discovery consequences from a specific training preference. Hosting and firewall rules also affect access. OpenAI crawler documentation.

Crawler preferences are not security controls for private files. Our private AI data-boundary guide explains why integrations, logging, backups, and access permissions matter alongside the model itself.

Measure visibility, accuracy, and business separately

As of October 2026, Google documents a dedicated Generative AI performance report, with worldwide rollout completed on August 31. It reports organic impressions in AI Overviews and AI Mode, with dimensions including pages, countries, devices, and dates. Its documented metrics do not include clicks or user queries. The data also appears in overall Web Performance, so the two totals should not be added together. Google’s AI performance report.

Bing’s AI Performance reporting measures citations. Its grounding phrases are generalized retrieval phrases, not full customer prompts, and citation counts are not clicks. Sampling and filtering also affect interpretation. Bing’s reporting guide.

OpenAI documents a ChatGPT referral parameter, utm_source=chatgpt.com. Preserve available referral information and connect it to qualified inquiries, while recognizing that people who never click remain unobserved in website analytics. OpenAI’s publisher FAQ.

Keep paid placements identifiable too. Buying an ad is a different activity from earning organic source selection. Our ChatGPT ads and SEO comparison explains attribution and incremental business in more detail.

Run a useful first experiment

Choose one important customer question and improve one substantial page. Add evidence you can stand behind, answer the question clearly, and keep a record of what changed. Where possible, retain similar unchanged pages for comparison.

Before and after publication, test a small fixed set of relevant questions and natural paraphrases across the engines you care about. Repeat them, record the date and settings, and save the answers and cited URLs. Include questions that do not name your business; supplying your URL tests comprehension rather than discovery.

Review whether the business is described accurately and whether each citation supports its surrounding claim. Track relevant search performance, referrals, and qualified inquiries alongside those observations. Allow time for recrawling and the buying cycle.

If your business appears in six of twenty sampled answers, report that result with its scope. It is not evidence of a 30% share of all AI searches. If inquiries are too sparse to distinguish a change from normal variation, keep the result inconclusive.

For a small firm, the first worthwhile investment is often a better explanation of the work it actually does. Build evidence into that explanation, make it easy to find, and connect the resulting interest to customers. SEO, AEO, and GEO can help organize that work when their labels remain tied to observable outcomes.

Looski helps firms build and operate AI workflows. If your next step is evaluating how an assistant should use your own business information, talk with us about the workflow.