Free GEO course: Get found by AI

Get recommended by ChatGPT, Gemini & Co.

10 short lessons for retailers and small businesses without prior knowledge — each with a concrete task for your website, so that AI assistants find, understand and recommend your company.

Samuel Welte, Software developer

Last updated: October 2026

After this course you will have …

a company sentence every AI understands — the same on your website, profiles and social media
a website that answers your customers’ 13 questions clearly and as text
your first case study with your own figures and an honest comparison online
your product data where AI assistants help people shop
Bing and Search Console set up — you see what the AI uses you for

10 lessons · 6–8 hours of work in total · free, no sign-up

For the first two lessons you need no access to your website — just an AI chat and a browser.

0 of 10 lessons done

Your progress is saved in this browser.

Frequently asked questions about the course

Who is this course for?: For retailers, online shops and small businesses that want to be found and recommended by ChatGPT, Gemini, Copilot and Perplexity — without prior knowledge and without an advertising budget. Every lesson ends with a concrete task for your own website.
What does it cost?: Nothing. The course is completely free, with no sign-up and no email address — and all the tools it recommends are free too: Bing Webmaster Tools, Search Console, Merchant Center, the AI assistants themselves.
Do I have to take the course “Get found on Google” first?: Not necessarily, but it helps: AI answers build on the normal search engines. If you already have Search Console, page titles and a Google Business Profile, you will get ahead faster here. Lesson 2 says exactly which foundations are needed — and the SEO course delivers them in nine short lessons.
What does GEO mean?: “Generative Engine Optimization” — making sure that AI assistants, which generate answers, find, understand and recommend your company. Some say AEO or “AI visibility”. The course needs none of these terms: it shows what you concretely change on your website.
When will I see results?: Product data and profiles often take effect within a few weeks; texts and mentions need months. Expect three to six months before the answers of the AI assistants change noticeably. That is why you record where you stand today in lesson 1 — only then does the progress become visible.
Is there a guarantee of being named in ChatGPT?: No — from nobody. OpenAI and Google themselves say that their systems select sources by relevance and reliability and that a mention can neither be bought nor forced. Anyone who promises you otherwise is selling you a trick. This course does what the providers themselves recommend.

How an AI recommends a company

To get started

What is this about?

What do I get out of it?

In five minutes you will understand how a customer question to ChatGPT turns into a recommendation — and why a pretty website is not enough for that. And you will record whether the AI already knows you today. Only then will you see later what has changed.

More and more people no longer ask Google but an AI assistant: ChatGPT, Google Gemini, Microsoft Copilot or Perplexity. Instead of ten blue links they get a finished answer — often with two or three companies the AI recommends. Whoever is not in there does not exist for that customer.

For an AI to recommend your company, four things have to happen: it has to find your website, understand what you stand for, trust you — and select you as the right solution for exactly this question. Every lesson in this course works on one of these four points.

The most important thing first: an AI searches differently from a person. From a question like “Which dealer in Vorarlberg sells e-bikes and has its own workshop?” it builds several sub-questions in the background — such as “e-bike specialist Vorarlberg”, “e-bike workshop Dornbirn”, “e-bike dealer Vorarlberg reviews” — and looks for sources for each of them. Google calls this query fan-out. Your website has to answer these sub-questions, not hit a single keyword.

And: the AI does not only read your website. It also reads what others write about you — customers, newspapers, directories, videos. That is why this course has two halves: lessons 2 to 8 work on your website and your data, lesson 9 on what can be found about you elsewhere.

Finally, to be honest: there is no guarantee. OpenAI itself says that ChatGPT selects its sources by relevance and reliability and that nobody can buy or force a placement. What you can do: make the AI's job so easy that it cannot help but understand you.

Beware of “GEO tricks”

This topic goes by many names — GEO, AEO, “LLM optimisation” — and by even more supposed secret tips: a file called llms.txt, paragraphs of exactly 50 words, a special “AI schema”, hidden commands in the page text. Google is clear about this: none of it is needed, and Google Search ignores llms.txt. What works is more fundamental — and it is in the next nine lessons.

The foundation: Google and Bing find you first

To get started

What is this about?

What do I get out of it?

AI assistants do not invent their sources — they take them from the normal search engines and from their own crawlers. If you are not properly indexed on Google and Bing, or accidentally lock out the AI crawlers, you are invisible to ChatGPT & Co., no matter how good your texts are. This lesson unlocks exactly that door.

Google says it explicitly: the AI answers in Google Search (“AI Overviews”, “AI Mode”) are built on the perfectly normal Google index. A page that is not indexed there cannot be cited there either. The same applies at Microsoft: Copilot and Bing's AI answers take their sources from the Bing index. And ChatGPT uses Bing results as well, alongside its own crawler. Bing has suddenly become important — even though hardly anyone searches there directly.

If you have done the course “Get found on Google”, half of it is already in place: Search Console, page titles, clean texts, images with alt text, internal links. This lesson adds three things that are not covered there: Bing, the robots.txt file and your host's bot protection.

The robots.txt is a small text file on every website that tells crawlers what they may read. OpenAI uses two different crawlers: OAI-SearchBot collects pages for search in ChatGPT; GPTBot collects texts for training future models. The two are independent — you can forbid training and still be found in ChatGPT. Many websites block both out of caution. That removes them from ChatGPT.

A second trap is even more common: hosts and services such as Cloudflare offer a switch “block AI bots”. It often blocks ChatGPT's search crawler too — regardless of what your robots.txt says.

No need to fear the technology

Nothing in this lesson requires you to program anything yourself. Where something needs changing, we do it for you — a is all it takes. For the check itself you only need a browser.

Your company sentence: say who you are in one sentence

To get started

What is this about?

What do I get out of it?

An AI can only recommend you if it can say in one sentence who you are, what you do and for whom. Today you write that sentence yourself — before the AI pieces it together from buzzwords. It is the foundation for every further lesson.

Microsoft and Google recommend the same thing for AI answers: clearly described “entities” — that is, unambiguously named companies, products and people — with facts that are understandable even without the rest of the page. Translated, this means: the AI must be able to read from your website: “[Company] is a [type of business] in [place], helps [target group] with [problem] and has proven experience doing so.”

Most websites say instead: “We are your partner for individual solutions with passion and quality.” A person skims over that. An AI can extract nothing from it — no business, no place, no service, no target group.

Just as important: consistency. Company name, description, industry, place and services should read the same on your website, Google Business Profile, Bing Places, social media and in directories. If your company appears sometimes as “Radhaus Rheintal”, sometimes as “Radhaus GmbH”, sometimes as “Bike Center Dornbirn”, to an AI those are three different companies — each with a third of your reputation.

And one more thing Google now explicitly recommends: a real name. Who stands behind the company, with experience and a photo. “The editorial team” or “your team” is nobody to an AI.

The 13 questions your website must answer

To get started

What is this about?

What do I get out of it?

An AI only recommends you if it can explain why you fit exactly this question. For that it needs answers — on prices, process, limits, results. This lesson shows the thirteen questions, checks your website against them and closes the biggest gaps. Along the way you answer exactly what your customers ask on the phone.

Think of your website not as a business card but as a knowledge base about your company: everything a new customer needs to know to decide in your favour is there — clearly, as text, in a place that can be found.

The thirteen questions an AI should find an answer to on your website: Who are you? What do you do? For whom? Which concrete problems do you solve? In which region? Which services or products exactly? How does it work? What does it cost, roughly? Who is it right for — and who is it not right for? Which results have you already achieved? Which alternatives are there? How do you differ? Which real examples prove it?

Most websites answer four of them. The gaps are almost always the same: price (“on request”), process, “not for whom”, results. But those are exactly the questions customers take to an AI — because they do not dare to ask them on the phone.

For retailers the questions sound a little different but stay the same: delivery time, shipping costs, returns, advice in store, repairs, which brands, which model for whom.

What it looks like

Example: Thirteen questions, checked at Radhaus Rheintal

Four customer questions and how the Radhaus website answers them before and after
Question	Before	After
What does it cost?	“Prices on request”	Service price list: 89 to 189 €, depending on drive and scope
How does it work?	not answered	Page “How your e-bike service works”: appointment, drop-off, check, call-back, collection within 48 hours
Not for whom?	not answered	“We do not repair road bikes or carbon frames — for those we recommend …”
Which results?	not answered	“2025: 1,140 services, 96 % of them finished within 48 hours”

Four paragraphs, one afternoon. Now an AI can give a well-founded answer to “e-bike service Dornbirn cost” and to “Who doesn't repair carbon?” — and name the Radhaus as its source.

One page, one topic — and the answer first

To get started

What is this about?

What do I get out of it?

AI assistants cite pages that answer a question clearly — with a title that says what it is about, a short answer right at the top and facts instead of adjectives. After this lesson you will have rebuilt your most important page exactly like that. The structure is no trick: it is the same one a hurried customer on a phone needs.

Microsoft states it as a rule for Bing and Copilot: One page, one clear topic. A page “Our services” covering sales, workshop, rental and accessories is not a good source for any of them. Every important service, every important product group gets its own page.

On that page the rule is: answer first, explanation second. Title and heading say literally what it is about. Directly below, two or three sentences give the most important answer and the key facts. After that, every subheading answers one sub-question — ideally in the customer's words. Large analyses by Semrush and Ahrefs show: pages with a clear summary, question-and-answer structure and clean sectioning are cited noticeably more often by AI answers. That is an observation, not a law — but one that fits what Google and Microsoft themselves recommend.

Add the things that build trust: concrete numbers (prices, times, quantities), a table wherever something is compared, a real example, name and date — and an FAQ with the questions from lesson 4.

What you do not need: paragraphs of a certain length, keywords in every line or a special “AI writing style”. Google says explicitly: there is no ideal text length and no obligation to chop content into small pieces. One section, one thought — no more rules than that.

Back up the old text first

You are about to rebuild a real page. Copy the old text into a document beforehand — then you can go back at any time without having to ask anyone.

What it looks like

Example: The structure of a citable page

The building blocks of the page “E-bike service and repair in Dornbirn” and their content
Building block	How it appears on the page
Page title	E-bike service & repair in Dornbirn — Radhaus Rheintal
Heading	E-bike service and repair in Dornbirn
Short answer	We service and repair e-bikes with all common drives (Bosch, Shimano, Yamaha) in our workshop in Dornbirn. A service costs 89 to 189 euros and is usually finished within 48 hours.
Subheadings	What does an e-bike service cost? · How long does the repair take? · Which brands do you repair? · What if my battery is getting weaker?
Table	Three service packages with contents and price
Example	“A commuter e-bike with 6,000 km: brakes, chain, software update — 149 euros, collected the next day.”
Author and date	Written by workshop manager Thomas Egger · last reviewed September 2026
FAQ	Six more customer questions from the workshop and the phone

Show what only you know: your own figures and real cases

To get started

What is this about?

What do I get out of it?

An AI has long known the general knowledge about your industry. What it needs from the web is information that exists only with you: your figures, your projects, your experience. That is exactly what turns “an article from the Radhaus” into the sentence “The Radhaus has proven experience with …”. After this lesson your first piece of evidence is online.

In 2026, in its guidance for AI search, Google said unusually clearly what counts: content from first-hand experience — your own tests, your own data, your own perspectives, things that cannot simply be summarised from existing pages. Google calls the opposite “commodity content”: the 400th article “7 tips for buying a bike” that any AI could write without you.

The scientific paper that coined the term GEO found the same: adding statistics, source citations and quotations increased visibility in AI answers the most — repeating keywords achieved nothing. The authors stress that the effect varies by topic. But the direction is clear: The easier a statement is to verify, the more likely an AI is to use it.

For a business this means: case studies. A good case study always has the same structure: customer or situation → starting point → previous approach → concrete problem → your solution → result in numbers → limitations. From that an AI can form the sentence: “This provider has experience with exactly this problem.” That is worth more than any advertising copy.

You have no projects but sell products? Then your evidence is: your own tests (“We compared twelve children's helmets on our test track”), your own figures (“These were our five most common reasons for complaints in 2025”), your own observations from the shop (“These are the three mistakes we see most often with e-bike batteries”).

Customer names only with permission

A case study with the customer’s real company name is the strongest — but only with their written consent. Without permission, describe the case anonymously (“a hotel with 24 rental bikes”). The figures stay the same, and they are what count.

Comparisons and honest decision aids

To get started

What is this about?

What do I get out of it?

People ask AI assistants above all when making decisions: Which is better? What does it cost? Is it worth it for me? Whoever answers these questions honestly — drawbacks included — is not only cited in the answers but named. After this lesson your first decision aid is online.

Semrush examined thousands of AI answers and found two things. First: in 62 % of the cases where a website was linked as a source, the company name did not appear in the answer — a “ghost citation”. Second: comparative content (“A or B?”, “the best … for …”) led to the brand being named around two and a half times as often as pure explanatory texts.

The reason is simple: whoever compares takes a position. An AI can build a recommendation from that: “For commuters with hills, Radhaus Rheintal recommends a mid-drive motor, because …” From an explanatory text “What is a mid-drive motor?” it can only extract knowledge — and it has that already.

The questions you should answer sound like this: “A or B?” · “What does … cost?” · “Is … worth it for me?” · “What are the drawbacks of …?” · “Do it yourself or have it done?” · “Who is … not right for?” For a retailer, for instance: hardtail or full-suspension? Bosch or Shimano? Buy or lease? Repair or buy new?

The decisive point: be honest. If your solution does not fit in one case, say so. Customers spot advertising in disguise immediately, and an AI prefers to cite the page that names both sides. Honesty here is not a virtue but the technique.

What it looks like

Example: Mid-drive or hub motor — an honest table

Comparison of mid-drive and rear hub motor by five criteria
Criterion	Mid-drive motor	Rear hub motor
Price (new bike)	from approx. 2,800 €	from approx. 1,900 €
Climbs	very good, uses the gears	good up to about 8 % gradient
Maintenance	chain and sprockets wear faster	low wear
Repair at our shop	yes, all common systems	Shimano and Bafang only
Ride feel	natural, balanced	noticeably pushes from behind

Below it, the recommendation by situation: “For commuters in the Rheintal with hilly routes: mid-drive motor. For flat terrain and a small budget: hub motor — which, however, you can get from us from only two manufacturers.” The last half-sentence costs no customer and brings credibility.

Product data is marketing now

To get started

What is this about?

What do I get out of it?

ChatGPT, Gemini and Copilot now show products directly in the answer — with price, image and availability. Which products those are is decided not by advertising but by the quality of your product data. For an online retailer this is currently the most concrete step in the whole course. You sell nothing online? Then skim this lesson and carry on with lesson 9.

More has happened here in 2026 than in all the years before. Google shows products from the Merchant Center in its AI answers and in the Gemini app — the free listings are enough for that, no advertising needed. With the Agentic Commerce Protocol, OpenAI has created a standard through which retailers deliver their product catalogue directly to ChatGPT. And Shopify has connected its catalogue directly: products from Shopify shops appear in ChatGPT, Copilot and Google's AI Mode without the retailer having to program anything. Perplexity runs its own merchant programme that understands the same data structure as Google Shopping.

What they all have in common: they read structured product data — name, description, images, price, availability, variants, item number (GTIN/EAN), reviews, shipping and returns. OpenAI says explicitly that this data helps ChatGPT understand products and present them correctly, and that the shopping results are sorted by relevance, not by advertising budget. If a field is missing, the product is missing — or it loses to the competitor's more complete one.

So something that used to be warehouse work becomes marketing: Clean product data decides whether an AI assistant suggests your product. The copied manufacturer text that reads the same in 40 shops loses to your own description with your photos and your experience from the shop floor.

What applies where — as of 2026

Google Merchant Center and Bing Merchant Center are open to retailers in Austria and Germany. The ChatGPT connection via Shopify is currently aimed mainly at shops serving customers in the USA; Perplexity’s merchant programme has so far targeted larger providers. That is changing fast — check the current state in the Shopify admin under “Sales channels” or with your shop system. The product data you clean up today applies to all channels at once.

What it looks like

Example: One product, two data records

The same children's helmet with incomplete and with complete product data
Field	Before	After
Title	Children’s helmet blue	Abus Youn-I 2.0 children’s bike helmet, size M (52–57 cm), blue
GTIN/EAN	missing	entered (13 digits from the manufacturer)
Description	Manufacturer text, identical to 40 other shops	Own text: “The helmet we recommend most often to parents of first-graders — thanks to the dial it fits over a hat too …”
Availability	not stated	In stock in Dornbirn, ships in 1–2 working days
Reviews	none	14 reviews, avg. 4.8, latest from 12 September 2026
Images	1 manufacturer image	4 own photos, one of them on a child

Both records describe the same helmet. Only the second can appear as the answer to “Which children's helmet fits over a hat?” — with price, image and stock level.

Let others talk about you

To get started

What is this about?

What do I get out of it?

An AI does not only ask: “What does the Radhaus say about the Radhaus?” It also asks: “What do others say?” An analysis of 75,000 brands shows: mentions on the web, videos on YouTube and brand awareness are closely linked to visibility in AI answers — the number of your own pages hardly at all. After this lesson you will have a plan for where your company should appear together with your topic.

Ahrefs examined 75,000 brands: most strongly linked to AI visibility were mentions on YouTube and on the web, searches for the brand name and links carrying the brand name as link text. The sheer number of own pages and the mere quantity of links played hardly any role. That is a correlation, not proof — but a clear one.

So the question is not “How do I get 200 links?” but: “On which trustworthy pages should my company appear together with my topic?” Context decides. “Radhaus Rheintal services rental e-bikes for hotels” in a trade publication is worth more than ten random links. You want to build a stable connection: Radhaus ↔ e-bike ↔ workshop ↔ Dornbirn — not Radhaus ↔ something to do with bicycles.

The places that count, depending on the industry: customer websites and references, partners and manufacturers (“find a dealer”), trade media and regional newspapers, associations and trade guilds, directories, YouTube, LinkedIn, podcasts and interviews, specialist forums and — used honestly — Reddit. For businesses with local customers, add Google Business Profile and Bing Places; Google and Microsoft both name them explicitly as sources for AI answers.

And a clear line: no purchased reviews, no invented forum posts, no artificial mentions. Google warns against it explicitly, Bing too. It gets noticed, it does harm — and it is unnecessary, because real mentions are easier to get than most people think.

Keep current and measure

To get started

What is this about?

What do I get out of it?

From now on you stop guessing. Bing shows you which questions its AI cites your pages for; Google shows how often you appear in AI answers; your visitor statistics show who comes from ChatGPT. And you repeat the five questions from lesson 1 — now with a baseline to compare against. You also learn what “current” really means.

Freshness counts. Ahrefs analysed around 17 million AI citations: on average, AI assistants cite noticeably younger pages than classic Google search, ChatGPT most of all. Microsoft names freshness explicitly as a factor. But “current” does not mean changing the date every week. It means: replacing old figures, adjusting prices, renewing photos, removing discontinued products, adding new results from projects, checking statements. An AI notices a changed date without changed content — and so does a customer.

Since 2026 there are real tools for measuring. Bing Webmaster Tools shows under “AI Performance” how often Copilot and Bing's AI cite your pages, which pages — and with which search phrases the AI found your content (“grounding queries”). That is the most valuable part: you see which topics an AI actually uses you for. Google Search Console has gained its own performance report for AI features (AI Overviews, AI Mode) showing how often your pages appeared there. And ChatGPT appends the tag utm_source=chatgpt.com to links in its answers — in your visitor statistics that shows you real ChatGPT visitors.

It pays to look at four things separately: cited (your page is linked as a source), named (your name is in the answer), recommended (you are suggested as a concrete option) and arrived (someone visits, writes, buys). The last one is the only one that counts — the first three are the way there.

Finally the priorities, in case time is short. Of 100 points of effort we would spread them like this: 25 on your own evidence and case studies (lesson 6), 20 on mentions beyond the website (lesson 9), 15 on clear pages and answers (lessons 4, 5, 7), 15 on the technical foundation (lesson 2), 10 on freshness, 10 on profiles and product data (lessons 3, 8) and 5 on technical details. Almost nothing on llms.txt, “AI schema” and other tricks. That is not a search engine's requirement but our weighting of the available evidence.

Patience — and the right yardstick

AI answers change from day to day and from assistant to assistant. A single answer proves nothing. Count trends over months: are you named more often? For more questions? Are more enquiries coming in that say “I found you in ChatGPT”? Those are the numbers that count.

Have we caught your interest?

Tell us briefly how things run in your business.
We will show you where you can save time and effort.

Get recommended by ChatGPT, Gemini & Co.

After this course you will have …

Frequently asked questions about the course

What is this about?

Open an AI assistant with web search

Ask five questions like a customer

Read the answer and note four things

Look at the competitors’ sources

Repeat in a second assistant

Record your starting point

Example: A bike shop asks ChatGPT

Common mistake: Asking for your own company name — and feeling reassured

Do this now

What is this about?

Check whether Google and Bing know you

Set up Bing Webmaster Tools

Read your robots.txt

If necessary: have the door opened

Check your host’s bot protection

Track down content no crawler can read

Switch on IndexNow, if available

Example: A robots.txt that forbids training and allows search

Common mistake: Locking out all AI crawlers across the board

Do this now

What is this about?

Write the company sentence

Do the buzzword test

Cross-check with an AI assistant

Put it on the home page

Align About us and Contact

Align your profiles

Make the sentence a habit

Example: The same business, two sentences

Common mistake: Three names for one company

Do this now

What is this about?

Copy out the thirteen questions

Go through your website like a stranger

Pick the three most expensive gaps

Attach each answer to the right page

Add questions from everyday life

The voice test

Example: Thirteen questions, checked at Radhaus Rheintal

Common mistake: Everything “on request”

Do this now

What is this about?

Choose your most important page

Title and heading: literal, not creative

The short answer in the first three sentences

Subheadings as customer questions

Numbers instead of adjectives

A table wherever something is compared

Readable address, author, date

FAQ from real questions

Example: The structure of a citable page

Common mistake: “Optimising” for the AI instead of writing for the customer

Do this now

What is this about?

Pick a case

Gather the figures

Write it in the fixed order

Attach evidence

Add the limitation

Third-party figures with a source

Publish and link

Example: A case study an AI can cite

Common mistake: Retelling general knowledge

Do this now

What is this about?

Collect your customers’ decision questions

Choose one question

Build the comparison table

Give every option its drawback

The recommendation by situation

Put costs in honest perspective

Sign off with the company sentence

Example: Mid-drive or hub motor — an honest table

Common mistake: The comparison your own product always wins

Do this now

What is this about?