Search behavior crossed a line. A growing share of buying questions — "best physio in Lisbon for runners", "affordable bookkeeping for a two-person café", "where can I get my brakes done tomorrow?" — are now asked to ChatGPT, Perplexity, Claude, or Google's AI Mode. The answer isn't a results page. It's a paragraph that names two or three businesses and moves on.
If you're named, you win the customer before they ever see a competitor. If you're not, you're invisible in a way classic SEO never was — there is no "page two" of an AI answer. The discipline of earning those mentions is called Generative Engine Optimization (GEO), or Answer Engine Optimization (AEO). Here's the playbook that actually works in 2026.
How AI engines pick who to recommend
Generative engines don't rank; they synthesize. When a user asks for a recommendation, the engine retrieves pages it can parse confidently, cross-checks facts across sources, and composes an answer from the businesses it understood best. That shifts what wins:
- Clarity beats keyword density. A page that plainly states "45-minute sports physiotherapy sessions, €60, open Saturdays, Alvalade" is quotable. A page of marketing prose is not.
- Structured data is read literally. schema.org markup (LocalBusiness, Product, Service, FAQPage) is the difference between the engine guessing your prices and knowing them.
- Corroboration matters. Engines trust facts that agree across your site, your Google Business Profile, review platforms, and directories. Inconsistent hours or addresses quietly disqualify you.
- Answer-shaped content gets cited. Pages that pose the customer's actual question as a heading and answer it in the first two sentences are exactly what retrieval selects for.
The GEO checklist for a small business
1. Make your facts machine-readable
Every page that matters should carry JSON-LD structured data: what you sell, prices, hours, location, and an FAQPage block answering the five questions customers actually ask. This one change moves you from "maybe parsed" to "reliably quoted."
2. Publish an llms.txt
Like robots.txt for the AI era: a plain-text file at your site root that gives language models a curated summary of your business and links to your key pages. It costs an afternoon and directly shapes how models describe you.
3. Write answers, not brochures
One page per real question — "how much does X cost", "X vs Y", "do you serve Z" — with the direct answer in the first sentence and detail after. This wins featured snippets in classic search and citations in AI answers with the same content.
4. Keep every surface consistent
Same name, hours, prices, and phone everywhere. AI engines cross-check; disagreement reads as unreliability.
5. Earn third-party mentions
Engines weight corroboration heavily: reviews on Google and industry platforms, local directory listings, a mention in a "best X in Y" roundup, a supplier or association page that links to you. You don't need a PR budget — you need five to ten independent places on the web that agree you exist, what you do, and that you're good at it. Ask your happiest customers for reviews and your professional network for the listings; it's the least glamorous and most effective GEO work there is.
6. Date your content and keep it current
Generative engines visibly prefer recent, dated material — a page updated this quarter beats an undated page from three years ago even when the content is similar. Put visible dates on your pages, refresh prices and hours the moment they change, and re-touch your key pages a few times a year. Freshness is a ranking signal you fully control.
A worked example: rewriting one page for GEO
Take a typical service page — a mobile dog groomer. The classic version reads: "Welcome to Pawfect Grooming! We're passionate about your furry friends and offer a wide range of grooming services tailored to your pet's needs." A language model retrieving that learns almost nothing it can quote: no prices, no area, no availability, no differentiator.
The GEO rewrite leads with retrievable facts: "Pawfect Grooming is a mobile dog groomer serving Lisbon and Cascais. A full groom (bath, cut, nails, ears) costs €45–€70 depending on size and takes about 90 minutes at your home. Book online — most appointments available within 3 days, including Saturdays." Then the page answers the questions people actually ask AI — "how much does mobile dog grooming cost in Lisbon?", "do mobile groomers handle anxious dogs?" — each as a heading with the direct answer in the first sentence, and an FAQPage schema block mirroring the same Q&As in machine-readable form.
Ask an AI engine either question, and only one of those two pages can be the source of the answer. That's the whole discipline in miniature: every claim on the page should be a fact an engine could safely repeat to a customer.
How to measure whether GEO is working
Classic SEO has rank trackers; GEO measurement is younger but very doable:
- Ask the engines yourself, monthly. Run your ten money questions ("best [your service] in [your city]", "how much does X cost near me") through ChatGPT, Perplexity, and Google AI Mode. Record whether you're named, what's said about you, and whether the facts are right. This takes twenty minutes and is your ground truth.
- Watch referral traffic from AI surfaces. Analytics now attribute visits from chatgpt.com, perplexity.ai, and AI Overview click-throughs. The volumes start small, but their conversion rates are typically far higher than classic search — these visitors arrive pre-recommended.
- Track "how did you hear about us?" The bluntest instrument and still the best. "ChatGPT told me" answers are already common in some verticals, and they'll tell you which engine matters in yours.
- Audit your facts in the answers. Being mentioned with wrong prices or dead hours is worse than absence. When you find an error, it almost always traces to an inconsistent or stale source you can fix.
The mistakes that quietly disqualify you
- Blocking AI crawlers reflexively. Some robots.txt templates block GPTBot, PerplexityBot, and friends wholesale. That may be a deliberate choice for a publisher; for a local business it's self-erasure. Decide consciously.
- JavaScript-only content. If your prices and services render only client-side in a heavy frontend, some retrieval systems see an empty page. Server-rendered or static HTML with the facts in the markup is the safe baseline.
- Keyword-stuffed answer pages. Engines synthesize meaning; repeating "best plumber Lisbon" fourteen times reads as noise, not relevance. One clear answer beats ten optimized paragraphs.
- Set-and-forget schema. Structured data that contradicts the visible page (old prices in the JSON-LD, new ones in the HTML) is worse than none — it's evidence you can't be trusted literally.
The catch: none of this works on a dead website. AI engines increasingly favor businesses whose sites are real, operating software — live catalogs, working booking, current prices — because those facts stay fresh and verifiable. A static brochure from 2023 loses to an operating storefront every time.
How the engines differ (and why you should care)
Treating "AI search" as one channel misses real differences in how each engine finds and cites businesses:
- ChatGPT blends trained knowledge with live browsing. For recommendation queries it leans on web search plus its sense of consensus — which is why third-party corroboration (reviews, roundups, directories) moves the needle here more than on-site tweaks alone. Its shopping and local answers increasingly pull structured product and business data, so schema pays off directly.
- Perplexity is retrieval-first and citation-heavy: every claim links to a source. It rewards exactly the answer-shaped page this guide describes — a clear question-heading with a quotable two-sentence answer is the unit of content it cites. Of all the engines, this is where a small, well-structured site can outperform a big, vague one fastest.
- Google AI Mode & AI Overviews inherit Google's index and local stack, which means your Google Business Profile is doing double duty: it feeds both the map pack and the AI answer. Hours, services, photos, and review velocity there matter as much as your website. Classic SEO hygiene transfers most directly here.
- Claude and assistant-embedded search (in messaging apps, phones, and cars) tend to answer from fewer sources with higher confidence thresholds — being consistently represented across the web matters more than being optimally represented in one place.
The practical takeaway isn't four separate strategies. It's that the checklist above is the common denominator all of them reward — and if your customer base skews toward one engine (Perplexity for technical buyers, Google for local walk-ins), you now know which lever to pull hardest.
The unfair advantage: a website that's already an app
This is where GEO stops being a marketing chore and becomes an infrastructure choice. Apps generated on Autoflowly ship as full operating businesses — real product catalogs, live booking calendars, working checkout — with clean semantic markup and consistent metadata by default. Your prices in the AI's answer are your actual prices, because the page and the database are the same system. And when the AI-referred customer arrives, an AI front desk answers them in seconds — which loops back into the reviews and mentions that feed the next recommendation.
GEO in 2026 rewards the same thing customers do: a business that visibly works. Get your facts structured, your answers direct, your surfaces consistent — and run on software that keeps all three true without you thinking about it.