MyAIDevelopment · Product · Operations notes

AI Search · Answer Engines · GEO/AEO

The biggest change in search in 2024–2026 is the growing number of screens where an AI-generated summary appears before the list of links. The industry calls the response to this GEO (Generative Engine Optimization) or AEO (Answer Engine Optimization).

Current State (as of 2026-09-28, with source dates)

Service Confirmed fact Source date
Google AI Overviews Available in more than 200 countries and territories and more than 40 languages Google announcement, 2025-05
Google AI Mode Five languages added, including Korean Google announcement, 2025-09
Google AI Mode Google announced it passed 1 billion monthly users one year after launch Google I/O, 2026-05
Search Console Generative AI performance report: launched 2026-06-03 for a subset of UK sites, expanded to all sites around 2026-08-31. Shows impressions, pages, countries, and devices only for AI Overviews, AI Mode, and similar features; no clicks, CTR, or queries. Data starts 2026-05-18 Google announcement 2026-06, expansion 2026-08
ChatGPT search Started providing answers with links to web sources OpenAI announcement, 2024-10
ChatGPT ads US test announced 2026-01. For Korea: plans announced 2026-05, pilot announced and reported 2026-06-19, and per OpenAI's notice launched on 2026-08-11 alongside the UK, Mexico, Brazil, and Japan. Shown only to logged-in adult Free and Go users; Plus, Pro, Business, Enterprise, and Edu have no ads 2026-01, 2026-06, 2026-08 OpenAI and press
Naver AI Briefing AI summary answers with source attribution introduced in integrated search Reported 2025-03
Naver AI Briefing Naver said on an earnings call it would roughly double the share of searches it covers by the end of 2026 Earnings call report, 2026-02

User and country counts are the providers' own announcements and change often. Re-check the originals before planning. Korean press (2026-06) and OpenAI's notice (2026-08-11) give different start dates for ChatGPT ads in Korea, so check OpenAI's advertiser guidance before running campaigns.

For consumer products aimed at Korea, check exposure in Naver AI Briefing with the same weight as Google AI Overviews. Naver keeps widening the range of searches where AI Briefing appears.

Clicks Can Fall

Pew Research Center analyzed March 2025 browsing data from 900 US adults (published 2025-07-22). On Google search pages with an AI summary, users clicked a result link 8% of the time; on pages without one, 15% of the time. Clicks on source links inside the AI summary were very rare.

This is the observation of a single study and may differ by industry and query type. The direction, however, is clear.

In 2024-02 Gartner predicted that "by 2026, traditional search engine volume will drop 25%". It is a prediction, not a measurement.

What Google Officially Says

Google Search Central's guidance (the 2025-05 blog post and the 2026-05 optimization guide) can be summarized as follows.

Caution: this is guidance for Google Search. For other answer engines, check each provider's documentation separately.

ChatGPT Search and Crawlers

According to OpenAI's documentation, crawler roles are separated.

User-agent Role If blocked
OAI-SearchBot Appearing in ChatGPT search results Hard to appear in ChatGPT search answers and snippets
GPTBot Collection for training generative models Signals that you do not want content used for training
ChatGPT-User Visits made on a user's request, such as asking ChatGPT to open a specific page OpenAI says robots.txt rules may not apply because the visit is user-initiated
OAI-AdsBot Visits landing pages submitted as ChatGPT ads to check ad-policy compliance and relevance If you run ChatGPT ads it is part of ad review, so decide together with ad operations

So it is possible to allow search visibility while refusing training collection. Decide robots.txt settings together with legal and content policy.

Practical Checklist

flowchart LR
    Q[Customer Questions] --> C[Unique Content]
    C --> T[Technical Access]
    T --> E[Entity Consistency]
    E --> M[Measure Mentions]
    M --> Q
  1. Build a question list: Collect questions customers might ask AI from interviews, sales, and support records.
  2. Create unique answers: Publish information only you know precisely, such as pricing, limits, comparisons, setup procedures, and real cases.
  3. Make it accessible: Check indexing, crawler permissions, rendering, and page speed.
  4. Stay consistent: Make sure product names, prices, and feature descriptions match across the website, docs, marketplaces, and review sites. AI answers combine many sources, so inconsistencies can turn into wrong answers.
  5. Measure: Use Search Console's generative AI report for the impression side. It has no clicks, CTR, or queries, so count the click side separately with landing-page referrers (chatgpt.com, perplexity.ai, and others) and UTM parameters. Add a manual check that regularly asks AI services key questions and records mentions and accuracy.

Bad Example / Good Example

Bad example:

Auto-generate 100 FAQ pages to rank well in AI, and ask community members to post praise for our product.

Good example:

For the real customer question "What happens to my data when I move from my current tool?", write a document that states the migration steps, limits, and time required precisely, and make the numbers on the pricing page match the docs.

The second action in the bad example is manufactured mention, which Google advises against; if it involves compensation, it also becomes an ad disclosure issue (document 09).

What Is Still Uncertain

References