For two decades, winning online meant ranking on page one. In 2026, a growing share of buyers never see page one. They open ChatGPT, Perplexity, or Gemini, ask a single question such as “what is the best payroll software for a 30-person company,” and act on the three names the assistant gives them. Google’s own AI Overviews now answer many searches above the results, and the buyer who reads the answer rarely scrolls to the links beneath it.
This shift has a name, or rather two: answer engine optimization (AEO), which is about getting your pages quoted in AI answers, and generative engine optimization (GEO), which is about getting your brand recommended in them. The two overlap with SEO but are not the same thing, and most small businesses have never measured either.
This guide walks through what changed, why your rankings do not protect you, and a checklist you can run this week without hiring anyone.
Why ranking first is no longer enough
A ranking describes a list. An AI assistant does not return a list; it returns a verdict with a few names in it. That verdict is composed from pages the assistant reads at query time plus what the underlying model learned in training. Review platforms, community threads, comparison pages, and “best of” lists are read heavily. A brand’s own homepage is read mostly for facts about that brand, rarely for the recommendation itself.
The practical consequence is uncomfortable: you can rank first for your main keyword and be absent from the answer, because the assistant built its shortlist from a review site you are not listed on and a competitor’s comparison page you never wrote.
Three things follow from this:
- Your website is one voice among many. What G2, Reddit, YouTube reviewers, and industry lists say about you often matters more to an assistant than what you say about yourself.
- Facts must agree everywhere. When your site says one price and a directory says another, the assistant hedges, and a hedging model usually leaves the uncertain name out.
- The page that answers the question wins, not the page that exists. A single clear comparison page beats fifty thin blog posts.
A new metric: share of answer
If you measure only rank and traffic, AI visibility work looks like it does nothing. The buyer who reads your name in an answer may not visit until they are ready to buy, and the buyer who never reads your name never visits at all.
The metric that captures this is share of answer: across a fixed set of questions your buyers actually ask, the percentage of AI answers in which you are named or cited, tracked per engine. Named and cited are different wins. Being named without being cited means the assistant knows you from elsewhere. Being cited without being named means you supplied a fact but not the recommendation.
You can measure a small version of this yourself in an afternoon.
The one-afternoon audit
Step 1: Write down 20 buyer questions
Mix four types:
- Category questions: “What is [your category] and how does it work?”
- Selection questions: “Best [category] for [your typical customer].”
- Comparison questions: “[You] vs [competitor]” and “[Market leader] alternatives.”
- Fact questions: “How much does [you] cost?” and “Does [you] integrate with [tool]?”
Step 2: Ask every engine
Run each question through ChatGPT (with search enabled), Perplexity, Gemini, and a Google search to see the AI Overview. For each answer, record whether you were named, whether a page of yours was cited, which competitors were named, every source cited, and any wrong fact about you.
Step 3: Count and sort
Compute your share of answer per engine. Then list every cited domain and sort by frequency. In most categories, five to eight domains account for the majority of citations. That list is the most useful thing you will produce all week, because it tells you exactly where to be.
The checklist: fixing what the audit finds
Make sure the assistants can read you
A surprising number of businesses blocked AI crawlers during the wave of concern about model training, using a blanket rule in robots.txt or a one-click CDN setting that also blocks the search crawlers assistants use to read pages today. Check for rules against agents such as OAI-SearchBot, PerplexityBot, and ClaudeBot. Blocking training bots is a legitimate choice; blocking search bots removes you from answers entirely.
Then check rendering. Open your pricing page, view the raw HTML source, and search for your actual price. If it is not there because a script adds it after load, most AI crawlers never see it. Server-side rendering for your key pages fixes this.
Write the three pages that get quoted
For commercial questions, assistants overwhelmingly cite comparison pages. Most small businesses have none. Write three:
- Your head-to-head against the competitor you lose to most.
- The category shortlist for your core customer type, with yourself honestly placed among real alternatives.
- The alternatives page for the market leader, which captures buyers who are leaving it.
Each should state its verdict in the first paragraph, use explicit criteria, include a table, admit where competitors are stronger, and end with a short FAQ. The page that wins every row gets treated as marketing; the page that tells the truth gets quoted.
Make the facts agree everywhere
Keep one sheet with the canonical version of your name, one-line description, founding year, location, current pricing and plan names, and integrations. Audit every profile against it: your review-site listings, Crunchbase, LinkedIn, app marketplaces, and any directories in your category. Add Organization schema to your site with a sameAs list pointing at all of those profiles, and publish a short llms.txt file that describes what you do in plain words. The goal is that a model checking your facts finds the same answer in every place it looks.
Get into the sources the assistants already read
Your source list from the audit is the target. If it is a review platform, claim and complete your profile and set up an honest flow for asking real customers to review you. If it is a community, have a named person from your team participate for real, disclosing affiliation and answering questions fully, including the ones that do not involve your product. If it is a publication’s “best of” list, pitch the editor with genuine differentiation. No purchased reviews, no fake accounts; the platforms detect it and the penalty is losing a source that feeds every answer in your category.
Correct what is wrong
The most damaging finding in most audits is not absence but error: an assistant quoting a price you changed two years ago, from a directory nobody updated. Search the wrong fact in quotation marks, find every page carrying it, and correct each: your own stale pages first, then profiles, then a polite email to third-party authors.
Measure again, and keep measuring
Rerun the same 20 questions monthly, or weekly if you are actively working on it. Share of answer only means something as a trend next to what you shipped. If you want the larger version of this exercise run for you, with a few hundred questions and every engine, a visibility audit from an AI search specialist such as aeogeolabs here will show you who the assistants recommend instead of you and what they appear to be reading.
What realistic progress looks like
Fact questions and comparison citations tend to move first, often within weeks, because assistants re-read your own pages quickly. Selection questions move as your presence on review sites and communities builds. Answers that come purely from a model’s training move last and can lag for months. Measured in share of answer rather than traffic, the trajectory is visible; measured in traffic alone, it is invisible until it suddenly shows up as inbound demos from buyers who already know your name.
The bottom line
The buyer’s path changed from “search, scan, compare” to “ask, read, act.” The businesses that notice are doing something unglamorous: checking that machines can read their sites, writing honest comparison pages, making their facts agree across the web, showing up where assistants look, and counting how often the answer is them. None of it requires a large budget. All of it requires starting with the measurement, because until you know which questions you are missing from, every other effort is a guess.