Generative Engine Optimization for Small Businesses: A Plain-Language Guide

Every few weeks a founder forwards me a proposal for "GEO services." The acronyms are new and the price assumes a marketing department. She has a laptop and a client at two o'clock. Generative engine optimization is real, but for a business of one to ten people it is mostly a short list of plain decisions, not a new discipline to outsource. The paper that coined the term, by Aggarwal and colleagues at Princeton and IIT Delhi, found that adding citations, quotations and statistics to a page could boost its visibility in generative engine answers by up to 40% across a benchmark of 10,000 queries. Show your evidence, be easy to read, and be consistent. That is most of it.

This is the "how" half of a pair; the companion article covers why AI search changes how clients find you. Here I will explain what GEO is in plain language, how AI answers pick their sources, which facts about your business have to match everywhere, what schema markup and JSON-LD actually do, what has changed with FAQ markup, whether you need an llms.txt file or crawler settings, and then give you a step-by-step checklist you can work through in an afternoon or hand to whoever manages your site.

What is generative engine optimization, in plain language?

The short answer: GEO is the practice of making your business easy for an AI answer to find, understand, trust and quote. It is what SEO becomes when the result is a paragraph instead of a list.

A generative engine is any system that searches, reads several pages and writes one answer with sources: ChatGPT with search on, Perplexity, Google's AI Overviews and AI Mode, Copilot. The Princeton paper describes the mechanism: the engine rewrites your question into several simpler queries, runs them through a search engine, summarizes each source, and writes a response grounded in those summaries. Google calls its version "query fan-out," and its Search Central documentation describes it as issuing multiple related searches across subtopics and data sources. Your page is never judged on one query. It is judged on whether it clearly answers one of the sub-questions the engine generated on the way to its answer.

Traditional SEO asked "can I rank for this phrase?" GEO asks "when a model breaks a client's question into pieces, is there a page of mine that answers one piece so cleanly it would be careless not to quote it?" Quotability is the new ranking. And quotability is a writing and positioning problem before it is a code problem. Brand clarity first, intelligent systems second: the order I use for automation is the order here too. Everything technical in this guide exists to make an already-clear position legible to a machine. If the position is not clear, the markup just labels the confusion accurately.

How do AI answers actually pick their sources?

The short answer: they retrieve pages that answer a sub-question directly, then prefer the ones that are readable, specific and backed by evidence. No engine publishes its full recipe, but Google, Microsoft and the academic work agree on the shape.

Microsoft is the most explicit. When Bing Webmaster Tools launched an AI Performance report in February 2026, showing site owners how often their pages are cited in Copilot and Bing's AI answers, its advice was short: strengthen depth and expertise, improve structure with "clear headings, tables, and FAQ sections," support claims with "examples, data, and cited sources," keep content fresh and accurate, and "align text, images, and video so they consistently represent the same entities." Google's guidance for AI Overviews points at the same fundamentals: crawlable pages, people-first content, internal links, accurate structured data. Neither company asks for anything exotic.

The Princeton experiments add texture. The largest gains came from pages that cited sources, quoted experts and included statistics, and the effect held on Perplexity, a live engine, where the improvement reached 37%. The authors note that what works varies by domain. The practical reading: write the way a careful expert writes when she expects to be quoted. Name the study. Give the number. Put the answer in the first two sentences under a heading that is the question. A page that hedges for three paragraphs before answering gives a model nothing to lift.

Which facts about your business need to match everywhere?

The short answer: your name, your business name, your role, your location, your category and your one-sentence position, identical across every profile a machine might read. In search language this is "entity consistency," and it is the cheapest GEO work there is.

An entity is the model's file on you; every page that mentions you either adds to it or contradicts it. Google's Organization structured data documentation is honest about why it exists: adding it to your home page "can help Google better understand your organization's administrative details and disambiguate your organization in search results." Disambiguate is the key word: there are other people with your name. The sameAs property in that markup lists your other official profiles, LinkedIn, Instagram, a directory, so a machine can connect them to the same file rather than treating them as strangers. Consistency is not about being everywhere. It is about being the same person everywhere you already are.

The audit takes about an hour. Open a spreadsheet with one row per place your business is described: homepage, about page, service pages, LinkedIn, Instagram bio, Google Business Profile, any directory, association, podcast or speaker page. Columns: name, business name, role, city, category, one-sentence position. Fill it in from what is actually written there. The discrepancies will surprise you. An old title on a podcast page, a city you left, three versions of the tagline. Fix what you control today and email the rest the exact sentence to paste. Every bio field on the internet is a vote for how a model will describe you, and some of your votes are still going to a version of you from 2022.

What is schema markup and JSON-LD, and how much do you actually need?

The short answer: schema markup is a short block of labeled facts in your page's code, JSON-LD is the format Google recommends for writing it, and a small business needs three or four types at most. It is a label on the box, not the contents.

Structured data is "a standardized format for providing information about a page and classifying the page content," as Google's introduction to structured data puts it, and Google "recommends using JSON-LD for structured data if your site's setup allows it, as it's the easiest solution for website owners to implement and maintain at scale." In practice JSON-LD is a script block in the page head that says, in machine-readable form: this page belongs to organization X, founded by Y, located in Z, offering services A and B, with official profiles at these addresses. Most website builders let you paste it into a header field, and Google's free Rich Results Test checks it.

For a coaching or consulting business the useful set is: Organization on the homepage with name, url, logo, address and sameAs links; Person on the about page for you as founder, with the same sameAs links; Service on each service page; and Article on each blog post with author and dates. The one rule that matters more than any property is Google's: "don't add structured data about information that is not visible to the user, even if the information is accurate." Markup describes the page. It does not replace it. If your about page does not state your city, write it on the page first, then label it. Ownership matters here too: keep a copy of your JSON-LD in a plain document you own, so the facts move with you if you change platforms.

Should you still add FAQ markup?

The short answer: write visible FAQ sections, yes; add FAQPage schema only if it is effortless, because Google retired the FAQ rich result in May 2026. The format still matters for AI answers. The code behind it has lost most of its payoff.

Google restricted FAQ rich results, the expandable questions under a search listing, to well-known government and health sites in 2023, and its Search Central changelog records a deprecation notice on May 8, 2026 stating the feature "will no longer appear in Google Search starting May 7, 2026," followed by removal of the documentation in June 2026. So the markup plugins have generated for years no longer earns a visual result in Google. Many guides have not caught up. Do not pay anyone to add FAQ markup as a headline deliverable.

What survives is the format. A visible question written the way a client would type it, followed by a direct answer, is exactly the chunk a generative engine wants to lift. Bing's guidance lists FAQ sections among the structures that "make content easier for AI systems to reference accurately." So keep a short FAQ at the end of every service page and post: answer first, explain second. The value was never in the code. It was in the discipline of answering a real question plainly. If your platform adds FAQPage markup automatically, leave it. The question people type is the heading. The first sentence is the answer. That format is the optimization.

Do you need an llms.txt file, and are AI crawlers allowed in?

The short answer: llms.txt is optional and unproven; crawler permissions are essential and often broken. Spend your ten minutes on the second one.

llms.txt is a proposal published by Jeremy Howard in September 2024 for a Markdown file at your site root that summarizes what the site is and lists its most useful pages for AI systems. It is tidy, cheap, and so far unread by the engines you care about. Google's John Mueller wrote in June 2025, as reported by Search Engine Roundtable, that "no AI system currently uses llms.txt," and that the consumer chatbots fetch normal pages for grounding but "none of them fetch the llms.txt file." My view: add one if your platform makes it a click, and expect nothing from it.

Crawler permissions are a different matter. Your site's robots.txt file tells automated visitors what they may read, and many security plugins and "block AI" toggles now write rules into it without asking you. OpenAI's crawler documentation explains that "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers," and that this setting is independent of GPTBot, the training crawler. You can decline training and still be recommended in search; Perplexity documents its crawler along the same lines. Check robots.txt this week: type your domain followed by /robots.txt and read it. If you see "Disallow: /" under a name you do not recognize, ask why. A surprising number of "why does AI never mention me" problems end at this file.

What is the step-by-step checklist for a small business?

The short answer: seven steps, in order, none of which needs a developer for more than an hour. Do them in sequence, because each one makes the next one cheaper.

1. Write the sentence. Who you serve, what you fix, how, in words a stranger would search. Everything below repeats it.
2. Run the entity audit. The spreadsheet from earlier. Bring every profile into line with the sentence. Email the owners of the ones you do not control.
3. Check crawlability. Read robots.txt. Confirm every service page and post is reachable through a plain link, not only after a click or scroll. Submit your sitemap in Google Search Console and Bing Webmaster Tools.
4. Add four schema types. Organization, Person, Service, Article. JSON-LD, matching the visible text, tested, copied into a document you own.
5. Restructure your top five pages. Heading as the question, answer in the first two sentences, evidence with a named source, a short FAQ at the end. Ignore FAQPage markup unless it is automatic.
6. Set a baseline. In fresh sessions, ask ChatGPT, Perplexity and Google AI Mode who the best [category] for [client] is, and what your name does. Record the answers and repeat monthly. In Bing Webmaster Tools, note which pages the AI Performance report shows as cited.
7. Corroborate. Once a quarter, put the sentence somewhere new you do not control: a guest article, a podcast bio, a partner page.

Notice what is missing: no new subscriptions, no forty generated pages, no daily posting schedule. The founders I see doing this well treat it as a system: set up once, documented in plain language, revisited monthly, living in accounts they own. If the checklist still feels like a lot, that is a time problem, not a competence problem; the positioning strategy work at The Scale Atelier starts with the sentence and includes the entity audit. If the fear of being left behind brought you here, I wrote about why standing still is not safer, and why that fear produces subscriptions rather than decisions.

Founder to founder: GEO sounds like something that happens to your website. Most of it happens to your sentences. Say who you are for, say it the same way everywhere, put the answer at the top of the page, and let the machines read it. Start with step one this week. The rest gets easier once the sentence exists.

Frequently Asked Questions

What is generative engine optimization (GEO)?

GEO is the practice of making your content easy for AI answer engines like ChatGPT, Perplexity and Google AI Mode to find, understand and cite. The term comes from a 2023 Princeton paper on what makes pages more likely to be cited.

Does schema markup help with AI search?

It helps machines confirm who you are and what a page is about, and Google recommends JSON-LD as the format. It does not substitute for clear visible content; Google asks that markup never describe information that is not on the page.

Does FAQ schema still work in 2026?

Google stopped showing FAQ rich results on May 7, 2026, so the markup no longer earns a visual result in Google Search. Visible question-and-answer sections still help AI systems lift accurate answers, so keep writing them.

Do I need an llms.txt file for ChatGPT?

No. Google has said no AI system was using the file as of mid-2025, and its own AI features do not require it. It is harmless to add, but crawler permissions and clear pages matter far more.

How do I check if ChatGPT can crawl my website?

Open yourdomain.com/robots.txt and look for rules that block OAI-SearchBot or all user agents. OpenAI states that sites opted out of OAI-SearchBot are not shown in ChatGPT search answers. Blocking GPTBot, the training crawler, does not affect search.

The Scale Atelier helps women-led coaching, consulting and expertise businesses define a position that clients and AI search can both repeat, then builds the documented systems behind it. Book a free 20-minute audit: no pitch, just a prioritized plan for your visibility and operations that is yours to keep either way.