AI Search Optimization, Explained From Zero

This page assumes you know nothing about SEO, old or new. No jargon debt, no homework. If you run a business and you have noticed that people increasingly ask an AI assistant instead of clicking through ten blue links, you already understand the problem this guide addresses. What follows is the plain-language version of what changed, the five ideas that carry almost all of the weight, a first-90-days plan you can start on a Tuesday, and the traps that catch beginners, usually at the exact moment they pull out a credit card.

One honest note before we start. This is what our agency does for a living, so we have an interest in you taking it seriously. Every claim below carries a source or a qualifier, and nothing here requires hiring anyone. A motivated owner can execute the entire beginner plan alone.

What changed about search, in three sentences

For twenty years, a search returned a ranked list of links and the contest was to be near the top of that list. Now, for a large share of queries, software reads a handful of pages and writes the answer itself, naming its sources as citations. BrightEdge's tracker measured Google AI Overviews appearing on roughly 48% of tracked queries by March 2026, and ChatGPT and Perplexity answer this way for every query they take.

That third sentence is the whole game. When an assistant composes the answer, being present means being one of the three to six sources it quotes or links. A business the assistant never retrieves, or retrieves and declines to use, is invisible no matter how good its website looks to a human. The mechanics of how those systems fetch and choose sources are covered step by step in our explainer on how AI search works; you do not need the full plumbing diagram to act on this page, but it helps to know it exists.

The discipline of earning those citations has an academic name: generative engine optimization, coined in a 2023 Princeton-led paper by Aggarwal and colleagues. Our what is GEO primer unpacks the term; here we will just use "AI SEO" because that is what people search for, and the two labels point at the same work.

The five ideas that carry everything

Almost every tactic you will ever read about reduces to one of five ideas. Learn these and you can evaluate any advice, any tool, and any sales pitch on your own.

Idea one: entities, or being a known thing

An entity is a thing the machine recognizes as distinct: a business, a person, a product, a place. AI systems keep an internal picture of the entities they know about, assembled from your website, your Google Business Profile, directories, reviews, and social profiles. When those records disagree, the picture blurs, and blurry entities get skipped or, worse, described wrongly.

One concrete action: write your business name, address, and phone number once, in one canonical form, then make every listing on the internet match it exactly. Same abbreviations, same suite number, same phone format. It is unglamorous, costs nothing, and it is the single cheapest confusion-remover available to a small business.

Idea two: retrieval, or being findable in fragments

Assistants do not read your whole site and form an impression. They issue searches behind the scenes, pull back the most relevant passages, and work from those fragments. Google's own documentation on AI features describes queries fanning out into multiple sub-questions, each retrieving its own best passage. If your key facts live in a paragraph that also contains your founding story, a joke, and a seasonal promotion, the fragment that gets retrieved is muddled.

One concrete action: for each important page, make the first paragraph under each heading a complete, self-contained answer to that heading. A reader who sees only that one paragraph, ripped from context, should still get accurate information with your business named in it.

Idea three: citations, or giving the machine a reason to quote you

Generative systems prefer sources that look verifiable. The Aggarwal paper tested nine content modifications against generative engines and found the strongest performers were adding statistics, quotations, and cited sources, reporting that these tactics "can boost source visibility by up to 40%" in generated responses. The same study found keyword stuffing, the classic old-SEO reflex, achieved little to nothing. Evidence formatting beat keyword repetition. That is worth sitting with, because it inverts a decade of folk wisdom.

One concrete action: take your most important page and add one dated statistic from a named source, one direct quote from a real person, and one link to an authoritative reference. Thirty minutes of work, and it moves the page toward the exact profile the research says generative engines favour.

Idea four: structure, or labelling your facts for machines

Structured data is a block of code on your page that states your facts in a format machines parse without guessing: this is a business, this is its name, this is its phone number. The vocabulary is maintained at Schema.org and the format is called JSON-LD. In our 12 Vectors methodology this is Vector 6, Structure, and it is the one genuinely technical item on the beginner list. Here is a minimal, real example for a local business; replace the values with your own and place it inside a script tag of type application/ld+json in your page's head:

{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "name": "Example Bakery",
  "url": "https://examplebakery.ca",
  "telephone": "+1-519-555-0134",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "12 Main Street",
    "addressLocality": "Brantford",
    "addressRegion": "ON",
    "addressCountry": "CA"
  },
  "sameAs": [
    "https://www.instagram.com/examplebakery"
  ]
}

Two honest caveats. Google states that structured data is not required for generative AI features, so treat schema as disambiguation insurance, not a ranking lever. And be suspicious of anyone selling "AI-specific" technical files: Google has publicly said it does not use llms.txt, a file some vendors still bill for.

One concrete action: add a LocalBusiness block like the one above to your homepage, then paste your URL into Google's Rich Results Test to confirm it parses.

Idea five: measurement, or checking what the machines actually say

Old SEO measured rank positions. AI SEO adds a second question: when an assistant answers a question in your category, does it mention you, and is what it says true? No dashboard does this perfectly yet, so beginners measure by hand, and hand measurement is genuinely enough to start.

One concrete action: write down the five questions a real customer would ask an assistant before choosing a business like yours. Once a month, ask all five to ChatGPT, Perplexity, Gemini, and Google, and record whether you were mentioned and what was claimed. That log becomes your baseline, and every change you make afterward gets judged against it.

Your first 90 days, if you run a small business

Sequenced so each month builds on the last. Budget a few hours a week, not a career change.

A timeline qualifier, because this is where expectations get abused: GEO and SEO timelines vary, and 3 to 6 months is a realistic window for measurable organic improvement on an established domain. New domains run longer. When we rebuilt the content library for Mattress Miracle, a Brantford retailer, the domain grew from roughly 1,000 to 82,400 monthly organic visits (SEMrush, April 2026), and that reflected sustained execution over many months, not a launch spike. Results depend on your industry, competition, and existing digital presence.

The traps that catch beginners

The cheap-package trap. The most expensive mistake in this field is usually the cheapest invoice. Packages priced at a few hundred dollars a month, or suspiciously round retainers with vague deliverables, tend to fund software subscriptions and status calls rather than actual research, writing, or engineering. We dissected one of these line by line in our $3,000 SEO package teardown; if you read one other page before spending money, make it that one. The tell is always the same: the invoice names activities ("optimization", "authority building") instead of artifacts you can open and inspect.

The volume trap. Publishing dozens of thin, near-identical pages used to be a shortcut and is now a documented penalty. Google's quality rater guidelines instruct raters to score scaled, low-effort, copied-or-paraphrased content at the Lowest rating, AI-generated or not. Ten pages with real substance beat two hundred without it, and the two hundred can drag down the ten.

The tool-worship trap. Most "AI SEO tools" are drafting aids. Useful, but a draft is not visibility. The machines reward what the Princeton researchers measured: evidence, clarity, and sources, none of which a tool generates on your behalf. As the paper puts it, traditional tricks like keyword loading offer "little to no improvement" in generative responses.

The guarantee trap. Nobody can promise a citation in ChatGPT or a spot in an AI Overview; the systems are closed and they change. What a provider can promise is labour and transparency about it. That is the logic behind our own Results Guarantee: if your existing domain shows no measurable organic search results after 12 months of work with Formative Digital, we work for free until you see them. Note what that is: a commitment to keep working, tied to an existing domain, with success defined in writing up front. Any pitch stronger than that deserves your skepticism.

Where to go deeper

When the beginner layer stops being enough, the next rungs are already written. What is GEO defines the discipline and its research base. How AI search works opens up the retrieval machinery this page only gestured at. AI search engine optimization is the full practitioner treatment of the tactics, and the primary-source teardowns in our research hub go deepest of all, including the rater-guidelines reading that explains what Google's classifiers are trained to reward. When you want the work done with you rather than by you, our services page shows how we structure engagements.

Written by Matt Griffin, founder of Formative Digital, Brantford, Ontario. Published 2026-07-20.