Generative Engine Optimization, Minus the Hype
What GEO actually changes for a small business site, what is repackaged SEO, and the seven-step checklist worth doing: readable HTML, open crawlers, one identity sentence, schema, llms.txt, earned mentions, and a free monthly tracking habit.
Published August 14, 2026
4 min read
PortMySite
Generative engine optimization is the name that stuck for a real shift: AI assistants answer with three to five named businesses where a search page offered ten links. Being one of the names is the new version of ranking, and being absent is the new page two.
Most of what is sold under the GEO label is technical SEO wearing a new badge, priced accordingly. This is the version without the badge: what actually differs, what you can check yourself, and the order to do it in.
What is genuinely different, in three sentences
AI crawlers read your raw HTML and mostly do not run JavaScript, so content that only appears after scripts run does not exist for them, even though Google can see it. Models assemble their opinion of you from many sources at once, so third-party mentions and consistent directory listings carry more weight than anything you say about yourself. And there is no rank tracker to hide behind yet, so measurement means asking the assistants real questions and reading what comes back.
Everything else, speed, structure, schema, good content, is the same work SEO already asked of you.
The checklist, in order
1. Confirm your content is in the HTML. View source, search for a sentence from the page. This is the gate everything else depends on, and it is the step builder-platform sites fail most. The five-minute visibility check covers it in detail, including the image-and-PDF trap. If your site fails here, fix this before touching anything below, because none of the rest matters to a crawler that receives an empty page.
2. Let the crawlers in, on purpose. Check robots.txt for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended. Blocking them is a legitimate policy some publishers choose; the failure mode is blocking them by accident through a default or a firewall rule. Make it an explicit decision either way. This site names each one with an explicit allow, so a future edit cannot silence them accidentally.
3. Write your one identity sentence, then repeat it everywhere. One accurate sentence: who you are, what you do, where. Put the same sentence on your site, your Google Business Profile, your directories, your social bios. Models cross-check sources, and disagreement between them reads as uncertainty; an assistant recommends the business it is certain about. Resist the urge to write a fresh clever version per platform. Consistency is the feature.
4. Give the machines structure. Schema.org markup is how you state facts about your business in a form models can lift directly: name, address, phone, prices, FAQs, services. If your pages already answer common questions, FAQ markup turns those answers into exactly the question-and-answer shape assistants reuse. Every page on this site carries it, which is not a coincidence.
5. Add llms.txt. A plain-text summary of your business at /llms.txt: what you do, pricing, key pages, honest limits. It is a convention, not a standard, and its cost is one page of writing. It is also the only file on your site written purely for the machine audience, so keep it factual and current, and let it say the uncomfortable things too. Ours states what cannot be migrated, because a model that quotes your limits accurately is building your credibility.
6. Earn mentions you do not control. Models weight third-party sources heavily: local press, industry writeups, directory profiles, community threads. This is the slowest item and the strongest one, and there is no shortcut that is not a scam. One genuine local article about your business outweighs a rewritten homepage. If you do outreach already, this is the same work with a second payoff.
7. Track it for free, monthly. Ask ChatGPT, Claude and Perplexity the questions your customers would ask, note who gets named, and ask the assistant why. A spreadsheet with a date column is enough. Paid trackers exist for brands monitoring hundreds of queries; a local business needs a calendar reminder.
The honest caveats
AI referral traffic is still small for most local businesses, whatever the vendor decks say, and we do not republish their statistics because we cannot check them. The reason to do this work now is not a surge that arrived; it is that the cost is close to zero when your site is built right, and structural fixes get expensive exactly when they become urgent.
The other caveat cuts the opposite way. If your site fails step one because the platform assembles pages in the browser, no checklist fixes that from inside the platform. That is a structural property of how the site is built, and it is the case where a rebuild is the fix rather than an optimization. The same rebuild is what moves the speed numbers, which is not a coincidence either: fast for a phone and readable for a crawler are the same property, a page whose content is simply there.
The short version
Readable HTML, open crawlers, one identity sentence everywhere, schema, llms.txt, earned mentions, and a monthly habit of asking the assistants. Four of those are a single day of work. One is an afternoon. One is slow and worth it. And the measurement costs nothing but a reminder.
That is generative engine optimization without the subscription. The parts that are hard are the parts that were always hard: building a site that serves its content plainly, and being a business other people write about.
Common questions
- What is generative engine optimization?
- GEO is making your business readable, identifiable and recommendable to AI assistants like ChatGPT, Claude and Perplexity, which answer with a short list of named businesses instead of ten links. In practice most of it is technical SEO under a new name: content in the raw HTML, crawlers allowed in, structured data, and a consistent identity across the web. The genuinely new parts are small: AI-specific crawler policies, llms.txt, and checking the assistants themselves.
- Is GEO different from SEO?
- About eighty percent overlaps. A fast, server-rendered, well-structured site with consistent business information is what both require. The differences that matter: AI crawlers mostly do not run JavaScript while Google does, earned mentions on third-party sites carry unusual weight because models cross-check sources against each other, and there is no rank-tracking industry yet, so you measure by asking the assistants directly.
- What is llms.txt?
- A plain-text file at yourdomain.com/llms.txt that summarises your site for AI models: what you do, your pricing, your key pages. It is an emerging convention rather than a standard, and no model is obliged to read it. It costs twenty minutes, cannot hurt, and is the one place you control exactly how your business is described to a machine. Ours is at portmysite.com/llms.txt if you want a working example.
- Do I need a GEO tool or agency?
- Not yet, for a small business. The checkable work is a day of effort you can do yourself, and the monthly measurement is asking three chatbots the questions your customers ask. The tracking tools are built for brands monitoring hundreds of queries across markets. Revisit that answer when AI referrals show up meaningfully in your own traffic, not when a vendor quotes an industry statistic.
- Why does my competitor show up in ChatGPT and I do not?
- Ask the assistant to explain its answer, which usually names sources. The common reasons are mechanical: their content is readable in raw HTML and yours loads through scripts, they are consistently described across directories and reviews while your listings disagree with each other, or they have third-party writeups and you only have self-published claims. Each of those points at a specific fix rather than a mystery.
Doing this yourself?
Or hand it to people who do it weekly.
Fixed price, 7-day delivery, full 301 redirect map, no downtime. Send the URL and get a real quote back within one business day.