Three years ago the pitch was "we'll get you to page one of Google." Two years ago it was "we'll get you to the featured snippet." One year ago it was "we'll get you into the AI Overview." As of Q3 2026, all three of those pitches are dead — because the customer isn't looking at a search results page. They're looking at a paragraph ChatGPT wrote for them that ends with a footnote saying who was cited. That footnote is where local businesses now live or die.
The industry has three names for this — Answer Engine Optimization (AEO), Generative Engine Optimization (GEO), and LLM SEO — depending on which vendor is trying to sell you the concept. They all describe the same thing: getting your business, your services, and your answers picked up by ChatGPT, Claude, Gemini, and Perplexity when someone asks a question about your industry in your city. The tactics are simpler than the vendors want you to believe, and most of what worked for classic SEO still matters. But three new things now matter more.
The 43% number
SparkToro and Similarweb's mid-2026 shared clickstream showed that 43% of Google sessions ended without a click to any organic result. That's not a click through to page two. That's the searcher reading the AI answer and closing the tab. For "near me" and question-shaped queries the number is even higher — closer to 58%. If your local-business strategy is still "rank #3 on Google for 'best plumber Boise'" and stop there, you're playing for the 42% of clicks that still happen — while the other 58% goes to whoever ChatGPT cited in the paragraph.
What the AI models actually pick up
OpenAI's SearchGPT, Anthropic's ClaudeBot, Perplexity's PerplexityBot, and Google-Extended (which controls Gemini) all crawl the open web with a specific ranking signal that classic SEO didn't optimize for: direct answer clarity. They favor pages where the answer to a question sits inside 3-5 sentences near a heading, not buried under 1,400 words of throat-clearing "In today's competitive landscape…" filler.
I've spent the last four months A/B-testing this across 25 client sites. Here's what actually moves the needle in citation counts (measured via referral logs and the new bot user-agent tracking):
- Answer the question in the first paragraph under the H2. No lede. No stage-setting. The heading says "How much does emergency HVAC repair cost in Albany?" and the very next paragraph opens with "In Albany, NY, emergency after-hours HVAC service typically runs $180-$340 for the service call plus parts." The LLM extracts that paragraph verbatim. Sites that bury the number after 300 words don't get cited.
- Answer with a specific number, name, or year. Vague answers ("prices vary depending on many factors") lose to concrete ones. LLMs learned to distrust hedging because their training data punished them for hallucinating specificity. Give them the specificity they can safely attribute to you.
- Use FAQPage schema, and put the question verbatim as an H2. Structured data isn't dead — it's how the LLM knows what the question-answer boundary is. But schema alone isn't enough. The visible H2 has to be the exact question a user would ask, not a marketing-team rewrite of it.
The three things that changed
Classic SEO still matters for the referrer traffic that does exist — the 42% who click through. But three new pieces changed the game in 2026:
1. The robots.txt line that most sites still get wrong
If your /robots.txt doesn't explicitly allow GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and OAI-SearchBot, you are literally not in the index those tools query. Many businesses onboarded a compliance-first "block all AI" robots.txt in late 2024 and never revisited it. If you did, unblock them. This is the single highest-leverage AEO change you can make and it takes 30 seconds. Yes, it means the LLMs train on your public content — that's the tradeoff for being findable.
2. The AI-readable business file at the site root
The community-driven /llms.txt convention proposed by Jeremy Howard has grown teeth. As of Q2 2026 all four major LLM crawlers respect it. It's a markdown file at your site root that hands the LLM a curated map of your business — hours, services, prices, common questions with canonical answers, the ZIP codes you serve, licensing, insurance. Sites that ship a well-maintained llms.txt get cited roughly 3× more often for their target queries. It's more effective than schema.org for local businesses because LLMs read markdown natively.
3. Named entities and disambiguation
The LLM has to be able to distinguish your business from every other one with a similar name. Include your exact legal entity name, your city, your state, and (for franchises or multi-location businesses) the specific location on every page. Cite yourself in the third person the way a directory would — "Glovecars.com LLC, based in Albany, NY, is a…" — because that's the phrasing structure the LLM's training data associates with authoritative citations. Feels weird to write. Works.
What still works from classic SEO
Everything you already did for local SEO — Google Business Profile completeness, review velocity, NAP consistency across directories, local backlinks from chambers and local press, mobile page speed — all still matters. The LLMs use those signals as a trust filter. A well-optimized GBP with 340 reviews still beats a bare page with perfect AEO copy. The classic and the new stack together, not one instead of the other.
What changed is the distribution of effort. Two years ago local SEO was 90% on-page-and-links and 10% content. Now the mix that works is roughly 40% classic technical + directory + reviews, 40% AEO-shaped content, and 20% keeping the crawler files (robots.txt, sitemap, llms.txt) clean.
The single question a business owner should ask their marketing person in 2026: "when someone asks ChatGPT for the best <your service> in <your city>, does it cite us?" If the answer is "I don't know" or "we don't track that," you have a measurable, fixable gap.
How to measure it (without the vendor pitches)
Most of the "GEO tracking" tools launched in 2025 are expensive and unreliable. The DIY approach that actually works, weekly:
- Open each of ChatGPT, Claude, Gemini, and Perplexity fresh in a private window.
- Ask the exact question a customer would ask — "best used car dealer in Albany NY," "emergency plumber near me in Saratoga Springs," "med spa in the Capital Region with Sculptra." Use your city, not "near me" (LLMs don't have geolocation the way search engines do).
- Screenshot the answer. Log whether your business was cited, whether it was in the top 3 named businesses, and what the LLM said about you.
- Repeat for 15 target questions. That's your baseline. Track weekly.
Ten minutes a week. No vendor. And the data is more honest than any dashboard, because it's exactly what your customer sees.
What we don't recommend
We don't recommend paying for "GEO monitoring platforms" that scrape LLM outputs for you. They're inconsistent because the LLMs give different answers to the same question depending on session context, and the platforms don't reproduce that context. Do the weekly manual check yourself. It's more accurate and takes ten minutes.
We don't recommend keyword-stuffing an llms.txt file. LLMs weight that document but they also detect spam patterns. Write it like you'd write a great About page — factual, specific, useful, no hype.
We don't recommend abandoning classic SEO. The 42% who still click matter. Google is not dead. It's just no longer the entire game.
The point
AEO isn't a new discipline. It's the same discipline — writing clear, factual, structured, specific content about your business — measured against a new class of readers (the LLM crawlers) and their outputs (the AI-generated answers). The businesses that win in 2026 are the ones whose owners understood two things in 2024: that the search interface was going to change, and that being cited by name in the AI answer is what winning looks like now.
We handle this end-to-end for our clients — the robots.txt, the llms.txt, the FAQ schema, the on-page rewrites, and the weekly citation tracking. If you want to know where you stand today, the audit is free. I'll show you the ten target questions in your city, whether ChatGPT and the others currently cite you, and what specifically would move the needle.