← Back to the knowledge hub

SEO & GEO:
the full overview.

How Google searches in 2026, what GEO really means, which AI crawlers exist and how to measure whether the work pays off. Backed by what Google, Bing, OpenAI, Anthropic and Perplexity document themselves – not by agency myths.

The short answer

SEO makes sure Google and Bing find and understand your pages and show them for matching searches. GEO (generative engine optimization) pursues the same goal for AI answers in Google AI Overviews, ChatGPT, Perplexity or Copilot: your page should appear there as a source. Google itself says optimising for its AI features is still SEO [2]. So the basis stays the same: crawlable technology, helpful content with first-hand experience, a well-kept business profile and clean measurement.

What's new are more crawlers, more interfaces and new reports. And a mention in an AI answer doesn't automatically bring a click. That's why this page sets out what the providers document themselves – and where agency promises begin.

How does SEO work in 2026?

Google describes its search in three stages: crawling, indexing and serving results [1]. During crawling, Googlebot downloads text, images and videos from pages it found via links or sitemaps. During indexing, Google analyses this content and stores it in a large database. Only when serving results is it decided which page matches a specific query.

You should know two statements from this documentation before you talk to an agency. First, not every page makes it through all three stages, and Google guarantees neither crawling nor indexing [1]. Second, Google doesn't accept money to crawl a website more often or rank it higher [1]. Anyone who "guarantees first place" is promising something Google itself rules out.

What changed in 2026?

The list of results is no longer the only interface. AI Overviews sit above many results, and in AI Mode Google answers questions in a dialogue. According to its own guide, Google uses its core ranking and quality systems for both [2]. The answers are produced via retrieval-augmented generation (RAG): the model fetches suitable pages from the search index and formulates an answer from them. On top comes the so-called query fan-out. In the background, the model runs several related searches at the same time to find further suitable sources [2].

For you this means: a page that isn't in the index can't appear in Google's AI answers either. Google puts the requirement like this: to appear as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown with a snippet in normal search [3]. According to Google there are no additional technical requirements [3].

And Bing?

Bing is often overlooked, even though Microsoft Copilot and the AI summaries in Bing draw on its index. Since February 2026 Bing has shown in Bing Webmaster Tools how often your pages are cited in these AI answers [13]. With the IndexNow protocol you can inform Bing and other search engines immediately when you publish, change or delete a URL. According to IndexNow, that doesn't guarantee immediate indexing [17]. We use IndexNow for tacobite.de too: with every change to the website we automatically resubmit all URLs from the sitemap.

Read more: best practice for websites →

What is GEO and how does it differ from SEO?

GEO stands for generative engine optimization. It means everything that makes generative AI systems find your content, understand it correctly and cite it as a source. The difference from SEO lies less in the technology than in the result. In SEO what counts is your position in the list of results. In GEO what counts is whether your page is cited in an answer, often alongside several other sources.

What Google itself says about GEO

In 2026 Google published its own guide for its generative AI features. The core statement: optimising for generative AI search is optimising for the search experience, and therefore still SEO [2]. From Google's point of view, terms like GEO or AEO (answer engine optimization) don't describe a discipline of their own. Google also lists what you don't need: no new machine-readable files or AI text files, no chopping your texts into small chunks and no special schema markup [2][3]. According to the guide, Google Search ignores an llms.txt [2].

Structured data still makes sense. It enables rich results and helps machines clearly assign company name, address or author. But it has to match the visible text [3]. According to Google it isn't a lever specifically for AI answers: it's not a requirement for generative AI search [2].

Where GEO goes beyond Google

ChatGPT, Claude and Perplexity run their own crawlers. The most important insight from their documentation: training and search run via separate crawlers. So you can prevent your content from going into model training and still appear in these services' search. The table summarises what the providers write themselves [4][5][6][7]:

CrawlerWhat it's for, according to the providerIf you block it via robots.txt
GPTBot (OpenAI)Content for training OpenAI modelsYour content shouldn't go into training. According to OpenAI, visibility in ChatGPT search depends on OAI-SearchBot.
OAI-SearchBot (OpenAI)Showing websites in ChatGPT's search featuresYour page no longer appears there as a search result, at most as a navigation link.
ChatGPT-User (OpenAI)Page fetches triggered by a person in the chatBecause people trigger the fetch, robots.txt may not apply here, according to OpenAI.
ClaudeBot (Anthropic)Content that may be used for trainingFuture content from your site is excluded from training.
Claude-SearchBot (Anthropic)Analysing content to improve search resultsAccording to Anthropic, may reduce your visibility in Claude's search results.
Claude-User (Anthropic)Page fetches when someone asks Claude somethingClaude can no longer fetch your page for such questions.
PerplexityBotShowing and linking websites in Perplexity results, not for trainingAffects your visibility in Perplexity search.
Perplexity-UserPage fetches when someone asks in PerplexityAccording to Perplexity, generally ignores robots.txt.
Google-ExtendedControls whether Google may use crawled content for Gemini trainingNo effect on inclusion in Google Search and not a ranking signal.

Our recommendation for most SMEs: allow search crawlers and decide consciously about training crawlers. Blocking GPTBot or ClaudeBot excludes you from training, not from search. Blocking OAI-SearchBot, Claude-SearchBot or PerplexityBot, on the other hand, costs you visibility there. Check your robots.txt even if you never added anything yourself. From our project experience, some hosting or firewall settings block bots across the board without anyone having decided that consciously.

Read the full article: SEO vs. GEO →

Which technical basis do Google and AI systems need?

Technology alone doesn't decide rankings. But it does decide whether your content gets a chance at all. Three questions are decisive: can a crawler fetch the page? May it go into the index? And does it work for people on a phone?

Crawling and indexing

The robots.txt controls which areas of a website crawlers may fetch. But it isn't a tool for keeping pages out of Google Search – Google says so explicitly [8]. That's what the noindex tag or password protection is for [8]. A typical mistake from our project experience: a noindex from the test phase simply stays in place after launch. An XML sitemap helps search engines find all important URLs. Whether a page is actually in the index is shown by URL inspection in Google Search Console.

Mobile-first indexing

Google uses the mobile version of your content for indexing and ranking [9]. Your mobile site should therefore show the same content as the desktop version [9]. If sections, images or structured data are missing on the phone, from Google's point of view they're missing entirely.

Core Web Vitals

Google measures user experience with three values, among others [10]. Largest Contentful Paint (LCP) stands for load time, Interaction to Next Paint (INP) for responsiveness and Cumulative Layout Shift (CLS) for visual stability. According to Google, good values are: LCP within 2.5 seconds, INP under 200 milliseconds and CLS under 0.1 [10]. Google writes that these aspects align with what its ranking systems seek to reward [10]. But good values alone won't get you to the top if the content doesn't answer the question.

Ten-minute tech check

  • robots.txt: open yourdomain.com/robots.txt. Is nothing important blocked, and are the search crawlers from the table above allowed?
  • Indexing: in Search Console under "Pages", check which URLs aren't indexed and why.
  • noindex: search the source code of your most important pages for "noindex".
  • Sitemap: submitted in Search Console and Bing Webmaster Tools, and without errors?
  • Mobile: is content missing on the phone that's visible on desktop?
  • Core Web Vitals: in the Search Console report of the same name, tackle the URL groups marked red first.
  • Bing: is the website verified in Bing Webmaster Tools? Only then will you see the AI citations there too.

Related: Website checklist before and after launch

Read the full article: AI crawlers and robots.txt →

How does local SEO work for SMEs?

For trades businesses, practices, retailers and service providers with a fixed catchment area, local search is often more important than the classic list of results. Anyone searching for "tax adviser Fellbach" or "car battery near me" usually first sees a map with entries from Google Maps. There your Google Business Profile decides, not your website alone.

How does Google rank locally?

Google names three main factors: relevance, distance and prominence [11]. Relevance means how well your profile matches the search. Complete and accurate details help with that. Distance describes how far your location is from the person searching or from the place searched for. Google derives prominence from signals such as links to your website and the number of your reviews [11]. And: according to Google, a better local ranking can't be requested, not even for payment [11].

What you can do concretely

  • Claim and verify the profile. Without verification you can hardly maintain anything [11].
  • Fill it in completely: category, address or service area, opening hours, services and attributes [11].
  • Add real photos and videos: team, premises, finished work [11].
  • Reply to reviews, including critical ones. Google explicitly recommends this [11].
  • The same contact details everywhere: website, Google profile, business directories and Bing Places. Bing advises local businesses to keep address, opening hours and contact details up to date there [13].

Reviews: what's allowed, what's prohibited?

Reviews work twice: on the ranking and on the decision of the people who see your profile. You may ask every customer for a review. According to Google's policies, incentives such as payments, discounts or free services in exchange for a review are prohibited [12]. Also prohibited are reviews with a conflict of interest, for example from your own employees, and selectively asking only satisfied customers [12]. Our tip from practice: after every completed job, ask for feedback with a direct link. And ask everyone, not just the fans.

Local visibility in AI answers

An AI answer to "good marketing agency near me" can only cite what's clearly and consistently stated on the web. If your opening hours on the website, in the Google profile and in Bing Places differ, you're not giving machines a clear fact. On your website, one page per location with real local content helps: projects on site, directions, contact person. Pages with merely swapped place names, on the other hand, help nobody.

Related: How we work on SEO, GEO and local SEO

Read the full article: local SEO for SMEs →

Which content do Google and AI cite?

In its AI guide Google says clearly what achieves more than any trick: unique, helpful content with its own perspective, for example first-hand [2]. Google calls this "non-commodity content", i.e. content that doesn't merely repeat what's already everywhere [2]. That's exactly where the opportunity lies for SMEs. You have experience from real jobs that no general guide has.

How do you recognise helpful content?

Google provides self-assessment questions for this [14]. Does the content provide new information, analysis or results? Was it written by someone who demonstrably knows the topic well? Is there a byline where you'd expect one? Among the E-E-A-T aspects (experience, expertise, authoritativeness, trustworthiness), Google says trust is the most important [14]. But E-E-A-T isn't a ranking factor of its own [14].

What Bing recommends for AI citations

In the announcement of its AI report, Bing names concrete points that favour citations [13]: clear thematic focus, headings, tables and FAQ sections. Plus evidence such as examples, data and sources, regular updates and consistency between text, image and video. That matches Google's advice to structure content with clear paragraphs and headings [2].

What doesn't help

Google warns against building a separate page for every variant of a query in order to influence rankings or AI answers. That violates the spam policy on scaled content abuse [2]. According to Google, many pages make a website neither higher quality nor more relevant [2]. Our rule of thumb: better one article a month that fully answers a real customer question than four superficial ones.

Example: guide or real answer?

Example (fictional): an electrical business from the region doesn't write "Photovoltaics: the big guide". It writes "What does a wallbox with installation cost in an old building?". In it, it names its typical price range and the cases in which it gets more expensive, and shows photos of real installations. Below the text it says who wrote it and when it was last updated. This page answers exactly the question a person asks – and that an AI model probably also searches for in the query fan-out.

Read more: blog & content for SEO and GEO →

How do you measure whether SEO and GEO work?

Without measurement, all SEO work remains a matter of feeling. The good news: the most important tools cost nothing. These four sources are enough for most SMEs:

ToolWhat you see thereWhich question it answers
Search Console: Performance reportClicks, impressions, click-through rate and average position [15]Which queries and pages bring visitors?
Search Console: report on generative AI featuresImpressions in AI Overviews and AI Mode, by page, device and country [16]Does your page appear in Google's AI answers?
Bing Webmaster Tools: AI PerformanceCitations in Copilot and Bing AI summaries, cited pages, grounding queries [13]For which questions does Microsoft's AI use your content?
Web analytics, e.g. Google AnalyticsVisits by source, enquiries, ordersDo search and AI actually bring customers?

What Search Console shows about AI search

Google counts clicks from AI Overviews and AI Mode in the normal Performance report under the search type "Web" [3]. Since 31 August 2026 there has also been a separate report for generative AI features, for all websites worldwide [16]. It shows impressions, i.e. how often your page appeared there. The report only appears once your website has collected enough impressions in these features [16].

Since 10 February 2026 Bing has shown the "AI Performance" report in Bing Webmaster Tools [13]. There you see how often your pages are named as a source in Copilot and in Bing's AI summaries. Particularly revealing are the grounding queries: search phrases the AI used when retrieving your content [13]. For ChatGPT and Perplexity, from our project experience, you're mainly left with web analytics. Visits with referrers such as chatgpt.com or perplexity.ai show that people click your links there.

From practice: Batteriecenter Ludwigsburg

How this works together is shown by our client project Batteriecenter Ludwigsburg (SEO and Google Shopping). The starting point before the start in April 2026: no visitor analysis, no tracking, no product feed in Google Shopping and content that wasn't aligned with search terms. From April to August 2026 visitors rose by 521 % and revenue by 162 % (average before/after). ChatGPT and other AI assistants also bring measurable visitors there. Without clean tracking, this development couldn't have been proven at all.

Which metric counts in the end?

Rankings and impressions are early indicators. What's decisive are enquiries, calls and orders. A mention in an AI answer without a click can still have an effect, for example when someone later searches directly for your company name. So also watch the queries containing your brand name in Search Console. If they rise, you're becoming better known, even if not every contact comes via a link.

Related: Batteriecenter Ludwigsburg case study

Read the full article: Google Search Console for SMEs →

When SEO and GEO aren't the right first step

SEO and GEO take time. According to Google, recrawling a changed page alone can take from a few days to several months [3]. If you need enquiries in the next four weeks, paid ads or direct outreach are more honest. If your website is barely usable on a phone, every optimisation fizzles out: then a relaunch comes first. And if nobody on the team has time to write down your customers' questions regularly, even the best content strategy stays on paper. We'd rather say that up front than sell a project afterwards that doesn't solve the actual problem.

If it fits: How we work on SEO & GEO. And if the basis is missing: Websites.

Frequently asked questions

Get found
instead of searching?

Let's spend 30 minutes on your visibility – on Google, in Maps and in AI answers. No ranking promises, no sales pressure.

Taco date or video call • Free • No strings attached