GEO 2026: definitive guide to rank in ChatGPT, Claude, Perplexity and Google AI Overviews

If you have been doing SEO for a while, you have probably noticed: more and more searches end in an AI-generated answer, not a list of links. ChatGPT, Claude, Perplexity, Google AI Overviews and Copilot are rewriting what it means to 'show up in results'. The game is not dead, but the rules are changing. We call this GEO: Generative Engine Optimization.

TL;DR

- Classic SEO optimizes for engines that return links. GEO optimizes for engines that return answers.

- What matters most: source authority, clear and structured content, schema.org, verifiable citations, and more recently llms.txt.

- Backlinking and technical SEO are still alive; content now also has to be 'citable' by an AI.

- New tools to measure it: Otterly.ai, HubSpot AI Search Grader, Answer Engine Insights, and Google Search Console AI reports.

What GEO actually is

GEO is the discipline of optimizing your digital presence so large generative models cite, recommend or use you as a source when they answer their users. You are not fighting for position #1 on a SERP: you are fighting to be part of the response generated on the fly.

It is a close cousin of traditional SEO, not a replacement. Authority, content quality and site architecture still matter. What changes is how your content is consumed: an automated system reads it, summarizes it, and repackages it for another human.

How ChatGPT, Claude and Perplexity work inside (simplified)

Each has nuances but the general pattern is:

1The user asks something.

2The system decides whether to search the web or answer from model knowledge alone.

3If it searches: fires one or several queries against its search engine (often Bing, sometimes proprietary) and pulls a handful of pages.

4Extracts relevant passages from those pages (chunking + embeddings + ranking).

5Builds an answer with those passages as context and, in many cases, attaches citations.

Perplexity is the most transparent because it shows sources prominently. ChatGPT and Claude also cite when browsing, though the UI is more subtle. For you as a brand, the goal is to land in those handful of retrieved pages and be clear enough that the model quotes you verbatim or mentions you.

Factors that drive being cited by an AI

After more than a year watching patterns on Perplexity and Google AI Overviews, the factors that correlate most with being cited are:

Domain authority: . Brands with good backlinks and historical presence still show up more. Classic SEO is not dead.

Content clarity: . Direct answers, lists, definitions. Models love content that already looks half like an answer.

Proper schema.org: . FAQPage, HowTo, Article and Organization help the engine understand what the content is about.

Verifiable citations and sources: . If you cite studies, data and experts, the model treats the page as more reliable.

Freshness: . Recent content tends to gain weight, especially in fast-moving topics like AI.

`llms.txt`: . A file at the root of the domain summarizing what content an AI can use and where to find the important stuff.

llms.txt: the new robots.txt

llms.txt is a proposal (not yet an ISO standard, but with real traction in 2025-2026) to talk to generative engines. It sits at the root of the domain, like robots.txt, and does two things:

1Tells AIs which parts of your site are most relevant and trustworthy.

2Provides a machine-readable summary of what your company does.

Basic structure: an H1 title with the site name, a summary, and sections with commented links to key resources. At sprintmarkt.com we use it to point to our services, portfolio, blog and pricing. It is short, precise and useful for a model that has five seconds to decide what to cite.

Useful schema.org for GEO

You do not need to implement the full schema.org catalog, but four or five well-placed types make a big difference:

Organization: name, logo, sameAs (socials), contact info. Mandatory.

LocalBusiness: if you have a physical presence. For local businesses in Valencia, Madrid or anywhere, this is GEO and local SEO at the same time.

Article: or BlogPosting on every post, with author, datePublished, dateModified and mainEntityOfPage. Helps the model grasp freshness and authorship.

FAQPage: for FAQ sections. Perplexity literally pulls Q&A straight from here.

HowTo: for step-by-step guides. Very citable in 'how to X' answers.

GEO-friendly content structure

If you want an AI to cite you, write with how it would read you in mind. Patterns that work:

TL;DR at the very top: . Three or four bullets with the answer. If the model finds a decent summary, your text is the one it uses.

Questions as H2s: . 'What is it', 'How it works', 'How much it costs'. Generative engines turn user questions into queries and look for pages with those headings.

Direct answers in the first paragraph of each section: . Skip the decorative intro; the model cuts chunks and if your answer is 400 words below the H2, it probably will not be retrieved.

Lists and tables: . Easy to parse, and models reproduce them almost verbatim.

Visible citations: . If you post data, link the source. The model learns you are a citer and treats you better.

Backlinks still matter

Even as the game shifts, links from media, universities and authoritative publications remain a strong signal. An article cited in El País, Xataka or a university is more likely to appear in ChatGPT answers than one on an isolated blog. Digital PR and collaborations with specialist media are part of modern GEO, not a separate discipline.

Need help with your project?

Calculate your budget in 2 minutes with our interactive tool.

Calculate budget

How we got SprintMarkt to show up in Perplexity (useful anecdote)

When we launched the Next.js rebuild, we did three specific things for GEO:

1Published a clear llms.txt with sections per service and links to pages with pricing.

2Rewrote the service pages with a question-based structure: 'What is included', 'How much it costs', 'How long it takes'.

3Implemented Organization, LocalBusiness (Valencia), FAQPage schema on key services and Article on every blog post.

In the following weeks, Perplexity started mentioning SprintMarkt in answers about 'web agency Valencia' and 'sprint app development'. No magic: just doing the homework with new criteria.

How to measure if an AI cites you

This is where the market is still immature. Tools we are using at SprintMarkt:

Otterly.ai: and similar: track brand mentions in ChatGPT, Perplexity, Gemini and Copilot answers with your defined queries.

HubSpot AI Search Grader: evaluates how generative engines see you and gives a score.

Answer Engine Insights: share-of-voice metrics in AI-generated answers.

Google Search Console (AI reports): Google has started showing clicks and impressions data in AI Overviews.

Manual monitoring: periodically running the ten most relevant queries for your business on each engine and tracking who shows up.

KPIs that start to make sense

Brand mentions in AI answers: per hundred queries in your sector.

Citation rate: share of times you are cited vs. mentioned without a link.

Position in citations: earlier citations tend to weigh more (similar to old position-based CTR).

Referral traffic from Perplexity and ChatGPT: . Still small for most sites, but growing.

Typical mistakes in GEO projects

Thinking GEO replaces SEO: . It does not. It is complementary. If your technical SEO is broken, GEO will not save you.

Keyword stuffing: . Generative models quickly detect content written for old-school search engines.

Ignoring authorship: . Pages without an identifiable author or an About trigger less model trust.

Not updating: . A 2022 post without dateModified ages faster than one with a recent date.

Practical plan to start this month

1Publish an llms.txt at your domain root.

2Implement Organization, LocalBusiness (if applicable), Article and FAQPage schema.

3Rewrite your three top-traffic pages with a question-based structure and TL;DR.

4Monitor your ten key queries on Perplexity and ChatGPT every two weeks.

5Land at least one backlink in an authoritative sector publication next quarter.

The 36 AI bots you should know in 2026

In 2026 there is no single 'AI bot'. There are at least 36 distinct crawlers feeding generative models, each with its own agenda. The strategic decision is binary per bot: allow or disallow in your robots.txt. The main ones today:

OpenAI: GPTBot (general crawl), OAI-SearchBot (ChatGPT Search), ChatGPT-User (user actions).

Anthropic: ClaudeBot, Claude-Web, anthropic-ai.

Google: Google-Extended (blocks Gemini training without blocking Googlebot search).

Apple: Applebot-Extended (distinct from classic Applebot).

Perplexity: PerplexityBot.

Meta: FacebookBot, meta-externalagent (Llama).

Common Crawl: CCBot — used by dozens of open source models.

Bytedance: Bytespider.

Cohere: cohere-ai.

Amazon: Amazonbot.

Huawei: PetalBot.

Our recommendation at SprintMarkt for a standard B2B company: allow all unless clear reasons not to (direct competition, premium paywalled content). If you want to appear in AI answers, blocking the crawler is the equivalent of putting a fence in front of your shop window.

Our own robots.txt explicitly allowlists all 36 and you can copy it as a template: [sprintmarkt.com/robots.txt](https://sprintmarkt.com/robots.txt).

What is happening with Google AI Overviews in 2026

Google AI Overviews (formerly Search Generative Experience) has moved from beta to massive rollout in Spain throughout 2026. The important parts for you:

Average click-through drops 30-40%: on queries where an AI Overview appears. People read the generated answer and do not click.

But the clicks that do come through convert 2-3x better: — they are users who read the answer and still wanted more. Less traffic, higher quality.

Being cited in the Overview matters more than classic #1: in many sectors. And the citations inside the Overview come from pages with clear structure, correct schemas, and domain authority.

Google Search Console reports AI Overviews impressions: since late 2025. Check monthly: if you show up there, double down on those topics.

Concrete tactic: identify your top 20 keywords, search each while logged into Google, note which trigger an AI Overview and who is cited. Those are your real GEO competitors, not the classic organic ranking ones.

GEO analytics tools: comparative table 2026

Until a year ago, measuring GEO was impossible. Today there is a small but functional market. Real options:

Tool	Strong in	Weakness	Approx. price
Otterly.ai	Tracking mentions in ChatGPT, Perplexity, Gemini	Young UI, basic dashboards	From 29€/month
Profound	Share of voice in LLMs, deep analytics	Expensive, enterprise-focused	From $499/month
HubSpot AI Search Grader	Free one-shot score	Not continuous tracking	Free
Peec AI	Multi-LLM monitoring with alerts	Very new startup, changing roadmap	From 49€/month
Rankscale	SERP tracking + AI Overviews	AI Overviews still beta in ES	From 39€/month
Manual with template	Free, full control	Takes 2-3h/month	0€

What we do at SprintMarkt: combination of Otterly.ai (continuous tracking on 10 key queries) + monthly manual monitoring with predefined prompts in ChatGPT, Claude and Perplexity. Total ~40 min/month, costs 29€/month, and we get proprietary data on how each client's presence evolves in each LLM.

If you do not want to pay, start with HubSpot's grader + a Google Sheet with 10 prompts you run every 2 weeks on the three main engines. With that alone you are ahead of 95% of the Spanish market.

Where we come in

At SprintMarkt we offer a combined SEO+GEO audit for 490€. It includes traditional technical review, GEO-friendly structure analysis, schema review, a proposed llms.txt, and a content plan designed to be cited. If you want your brand to stay visible when half of searches go through an AI, it is a good first step. Drop us a line and we will put it together.