Reading Time: 3 minutes

To correctly optimise for Google’s AI Overviews, we need to know how they are created

Not all AIs are created equal, and what works for Google AI might not be as effective for ChatGPT. It is all about brand (Google is in the Search game) and advancements in AI technology.

In short: AI Overviews are created = query type → Information retrieval → LLM Synthesis → Summary (+ citations).

AI Summary

Google’s AI Overviews are built through a five-step process. Informational queries dominate, using sources like Google’s index, Knowledge Graph, feeds, forums, and reviews. Gemini rephrases results into plain-English overviews, with citations. Unlike ChatGPT or Grok, Google’s system favours authority, visibility, and search bias while retaining blog relevance. Summarise in ChatGPT

 

How Information Is Created in Google AI Overviews

Step	Description
1. Query type	Informational (e.g., “how to” questions) or transactional intent (e.g., “near me”, “buy now”). Overviews are not available or relevant for navigational intent.
2. Information retrieval	Data is gathered from Google’s Search index, the Knowledge Graph, product feeds, forums, and review sites.
3. LLM synthesis	Retrieved data is passed through an adapted Gemini large language model (LLM) to generate the overview shown in the search result. Other variables like location matter.
4. Summary	Collated data is rephrased, condensed, and structured into a plain-English description (the “overview”). Errors can occur due to limitations in the LLM’s reasoning and context handling.
5. Citations	Google adds links to supporting websites that influenced the overview. These are accessible by clicking the link icon and then selecting the source website.

1. Query type

• Informational Intent: This is where current AI optimisation thinking is flawed. No advanced AI system wants to rely on website data, eg blogs, when answering an informational query (think question). This is where trying to optimise a current “how-to” blog is pointless. Information is the remit of AI; they have the knowledge. However, not all AI modes are advanced, and Google AI Overviews use a cut-down version of Gemini, which, in a sense, has to show a search bias, as it is Google after all.

Fortunately for Google AI overviews, there is still life left in all of those millions of blogs that answer a question. If you want to appear top of Google, then read on; if you are more concerned with pure AI systems, then move on.

AI Comparison – “How to climb a mountain?”

System	Example Answer	Information retrieval
Google AI	How to climb a mountain? Takes data from WikiHow, Red Bull, and YouTube.	Explicit links to curated sources
ChatGPT	How to climb a mountain? Gives structured guidance without citations. States: “I relied on my general training.”	No references unless asked; relies on trained knowledge
Gemini	How to climb a mountain? Supplies a single concise answer. States: “The response was based on general knowledge.”	General knowledge, no citations
Perplexity	How to climb a mountain? Uses data from Much Better Adventures, YouTube, and Reddit.	Always cites multiple external sources
Grok	How to climb a mountain? Provides a general knowledge answer without citations. States: “My response was based on general knowledge.”	No sources, relies on general knowledge
Deepseek	How to climb a mountain? Provides a general knowledge answer with book citations. States: “I did not rely on a single source, but rather synthesised information from a wide range of established, authoritative resources which are listed as book references, not websites. Weird eh!”	Book sources, relies on general knowledge
Summary: Google AI & Perplexity act more like search engines with citations, while ChatGPT, Gemini & Grok act more like assistants synthesising knowledge.

Where AI systems are weak

• Transactional Intent: This is where you can win big. Here, AI systems are weak. Knowledge avails you nothing if you don’t put any of it into practice, and of course, AI mode is data retrieval, not experience-led. AI has no skills, just knowledge.

AI Comparison – “Recommend me a London plumber for a boiler service?”

AI / Engine	Prompt	Sources	Results
Google AI	“Recommend me a London plumber for a boiler service”	Search and Google’s Knowledge Graph	Happy Dog Plumbing Plumb London Pimlico Plumbers (#1 in Google)
ChatGPT	“Recommend me a London plumber for a boiler service”	Google Maps / Google Business Profiles – Google’s local business database	A&C Heating and Plumbing Ltd Capital Boilers North London Plumb London
Gemini	“Recommend me a London plumber for a boiler service”	Search, Company websites, Third-party review sites	WPJ Heating Plumb London Homecure Plumbers Pimlico Plumbers
Perplexity	“Recommend me a London plumber for a boiler service”	Google Reviews, Perplexity Search, Trustatrader, Trustpilot, Bestlocalrated.co.uk, Yelp, Checkatrade	Heat Knights Vaillant Boiler Service PCH Boiler Repair City Plumbers London
Grok	“Recommend me a London plumber for a boiler service”	Google, Checkatrade, posts on X	Pimlico Plumbers Homecure Plumbers Boiler Engineer 4U RARE Plumbing
Deepseek	“Recommend me a London plumber for a boiler service”	Training data, scraped text from the web	Pimlico Plumbers ADEY Professional Heating Greenwich Plumbing Ideal Plumbing & Heating British Gas

👉 The overlap (e.g. Plumb London, Pimlico) shows trusted industry leaders, while differences reveal each AI’s data sources and biases.

 

Are Google’s AI Overviews here to stay?

Probably, although no doubt they will move under the adverts and be cut down in size. Also, links will be added to the content with click-through links, rather than the current format. Google will forever push users into its AI mode. Perhaps new keyboard shortcuts will result in faster scrolling through overviews.

However, Google has recognised that Gemini was rushed out in response to Sam Altman’s OpenAI/ChatGPT, Grok and DeepSeek and contains a number of embarrassing errors. Google AI overviews are created by a cut-down Gemini, which actually makes even more errors.

Google Rankings Can Guarantee ChatGPT Visibility, but only to subscribers

The rest of us mere mortals using ChatGPT have to put up with Bing results.

 

 

Citations

1. Google AI Overview Vs. Bard/Gemini.
2. GPT-5, the smartest, fastest, and most useful model yet (or is it!)
3. AI Search Optimisation in 2025: Insights from 41M Results Across ChatGPT and Google AI.
4. Read more on Gemini

Takeaway: Specialised GEO expertise like Circus AI delivers faster, measurable results than in-house teams in the UK.

Citations for a more comprehensive answer

General Queries

• What process does Google use for AI Overviews? Google AI Overview vs Gemini

• Steps behind Google AI Overviews explained. Search Engine Journal – How Google AI Overviews Work

• How does Google AI generate search summaries? Google – AI Overviews Help Guide

Comparative Queries

• Are Google AI Overviews better than Perplexity results? Perplexity.ai Blog

• Why does Google AI cite sources but ChatGPT doesn’t? OpenAI – How ChatGPT Works

SEO & Optimisation Queries

• Do blogs still rank in Google AI Overviews? Blogging & SEO in AI Search

• SEO strategy for Google AI Overviews in 2025.  Vanilla Circus – AI Search Optimisation