Course lesson · Beginner
How AI Changed Discovery
AI changed discovery by moving the comparison step from the buyer to the machine: instead of scanning a list of links, people increasingly read an AI-written answer that names a few options and cites some sources. That shifts what marketers compete for, from a click on a ranked link to being included, described accurately and cited in the answer. This lesson explains where AI answers appear, how they're built and what that means for your work.
In this lesson
- Where people now meet AI answers
- Training versus retrieval
- What's different for brands
- How to adjust your thinking
- Exercise and quiz
Where do people encounter AI answers?
| Surface | Examples | How answers are built |
|---|---|---|
| AI features in search engines | Google AI Overviews and AI Mode | Built on Search systems; links to supporting pages (Google) |
| Chat assistants with search | ChatGPT search, Perplexity, Copilot | Fetch web results and cite them |
| Chat assistants without search | A model answering from training | No live sources; knowledge has a cutoff |
| AI inside other tools | Browsers, office apps, phones | Varies by product |
A single buyer might use several of these during one purchase.
What's the difference between training and retrieval?
Two broad mechanisms shape what an AI says:
- Training: the model learned patterns from large amounts of text before release. What it "knows" about your brand from training can be out of date.
- Retrieval: at answer time, the system fetches documents, for example from a search index, and writes an answer grounded in them. Researchers call this retrieval-augmented generation (Lewis et al.).
| Training | Retrieval | |
|---|---|---|
| When it happens | Before release | At question time |
| Freshness | Fixed at a cutoff | Can be current |
| Shows sources? | Usually not | Often yes |
| What you can influence | Long-term: being widely and accurately described | Near-term: crawlable, clear, relevant pages |
Crawler controls reflect this split. OpenAI, for example, uses OAI-SearchBot for search and GPTBot for training-related crawling, controlled separately (OpenAI).
What's different for brands?
| Classic search | AI answers |
|---|---|
| Ten links; user compares | A few named options; AI compares |
| Position 1–10 | Included or not; recommended or just mentioned |
| Your page's title and snippet | The AI's description of you, in its words |
| Clicks as the main outcome | Influence can happen with no click |
| Same results for most users | Answers vary between runs and users |
The last two rows matter for measurement. A buyer might read an answer recommending you, then search your brand later, so analytics credits "branded search", not AI. And because answers vary, you measure with samples rather than a single rank check.
What stays the same?
Plenty. AI search still needs crawlable pages, clear information and a good reputation. Google says there are no additional technical requirements for AI features beyond being indexed and eligible for a snippet (Google). Content that's genuinely helpful, a core idea in Google's guidance (Google), remains the best foundation.
A quick example
Illustrative. A marketing manager needs a webinar platform. In the past, she'd search, open five review articles and build a spreadsheet. Now she asks an assistant: "Which webinar platform suits a 50-person B2B company that needs HubSpot integration?" The answer names three platforms with one-line reasons and links two review sites. She shortlists those three. A fourth platform, arguably a better fit, isn't mentioned, because its HubSpot integration is only described in a PDF, and review sites cover an older version.
For that fourth brand, the problem isn't ranking. It's that the facts an AI needs aren't easy to find.
How should you adjust your thinking?
- Think in questions, not keywords. Buyers describe their situation in full sentences.
- Make facts easy to find. State pricing, integrations and fit clearly, on crawlable pages.
- Care about what others say. Reviews, comparisons and community posts feed AI answers.
- Measure inclusion and accuracy. Ask: are we named, recommended and described correctly?
What's still uncertain?
AI products change quickly, and companies don't publish how they choose which brands to name. Be wary of anyone claiming a guaranteed formula. This course focuses on documented mechanics and on measuring your own results.
Exercise
Run 5 category questions in ChatGPT and in Google, and note which sources each shows. Use realistic buyer questions for your category. For each, record which brands are named, which sources are linked, and whether your brand appears. Note any differences between the two.
Quiz
- What's the main shift AI brings to discovery?
a) Ads disappear b) The AI does more of the comparing c) Websites stop mattering Answer: b
- Retrieval means:
a) Fetching documents at answer time b) Training a new model c) Deleting old pages Answer: a
- Which OpenAI crawler is used for search?
a) GPTBot b) OAI-SearchBot c) Googlebot Answer: b
- Why might analytics undercount AI influence?
a) Buyers may later visit via branded search or direct b) AI blocks analytics c) Google hides all data Answer: a
- Does Google list special technical requirements for AI features?
a) Yes, a special schema b) No additional requirements c) Yes, a paid listing Answer: b
Next: Lesson 3: SEO basics you still need
Sources
- Google Search Central, AI features and your website
- Google Search Central, Creating helpful, reliable, people-first content
- Lewis et al., Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
- OpenAI, Overview of OpenAI crawlers
Questions
Do people still use Google?
Yes. AI features now sit inside Google Search, and many people use both search engines and AI assistants.
What is retrieval in AI answers?
It's when an AI system fetches current documents, for example from a search index, and uses them to write its answer.
Does every AI answer link to sources?
No. Some surfaces show sources prominently; others answer from what the model learned in training, with no links.