An Overview of GEO / AI Search in 2026

January 19, 2026


Nearly a year ago, in March 2025, I wrote a blog post about AI search. At that time, AI search optimization, now often called GEO (generative engine optimization) was a fairly new phenomenon.

Now, almost a year later, I put together a presentation for one of my HBS courses on the state of AI search. Here’s what I learned along the way:

The Current Context: Pressure and Confusion

AI Search Context - Marketers under pressure

Marketers are under extreme pressure to understand and adapt to AI search, but separating truth from rumor is genuinely difficult.

Traditional search is undoubtedly under pressure from AI summaries and “zero click” answers. According to Bain research, about 80% of search users rely on AI summaries at least 40% of the time.

That said, headlines like “Panic! at the GEO” and “AI search has brands scrambling” are creating a sense of urgency that may be disproportionate to the actual impact.

Marketers being pushed to act without clear direction

This pressure to take action is often top-down. Executive teams feel pressure from boards and investors to “adopt AI” and have an “AI strategy,” but often lack clarity on the specifics.

CMOs are already some of the shortest-lived executives across industries (particularly in tech), and inaction can come at the cost of a job.

As a result, buying a tool - regardless of its actual utility — creates the appearance of doing something about AI search, which can ease the pressure on teams to demonstrate adaptability.

Putting AI Search in Perspective

AI search remains small but growing

An important reality check: AI search remains small today.

Globally, AI platforms currently drive just <1% of all internet traffic, a tiny fraction compared to the 48.5% from organic search. Google still sends 300 times more traffic to websites than all AI platforms combined.

That said, the trajectory matters. Like SEO before it, AI search is likely to grow, and ranking systems will mature over time.

We’ve seen this pattern before: the 1990s saw experimentation and learning, 1998 brought PageRank (and spammy tactics like keyword stuffing), the 2000s saw the SEO industry mature with ranking evolving to balance factors like content quality, and the 2010s shifted toward E-E-A-T with algorithm updates punishing spammy tactics.

Three factors of AI search: Training Data, Web Search, Tooling

A comprehensive view of AI search considers three distinct factors:

  1. Inclusion in Training Data — Is your brand included in LLM training data, in a positive light?
  2. Appearance in Web Search / RAG — Is your brand visible and your website cited in web searches and overviews?
  3. Availability of Owned Tooling — Do you have branded tooling available where possible (e.g., ChatGPT Apps, MCP servers)?

Most GEO tools today are focused primarily on factor #2 (web search/RAG), but a complete strategy should address all three.

(I have a longstanding interest in MCP servers, so am biased; many marketers would not include #3, but I think it’s importnat!)

Factor 1: Training Data

Training data importance and tactics

Appearance in training data is important but difficult to track or directly influence.

To use the data we have, GPT-3 was trained on: Wikipedia (3GB, 3% weight), WebText2 (20GB, 21% weight), Books1 (12GB, 8% weight), Common Crawl filtered (400GB, 60% weight), and Books2 (55GB, 8% weight).

Ways to take action:

The key insight here is that positive mentions on user-generated content sites carry significant weight.

Factor 2: Web Search / RAG

Being #1 matters less, multiple mentions matter more

In the web search layer, being #1 matters less than it used to.

AI Overviews don’t just cite the “first blue link”—67.82% of AI Overview citations don’t rank in Google’s top 10.

The implication is that you should aim for multiple positive mentions across a range of sites through tactics like PR placements and backlinks, and ensure your brand’s info and name appears near relevant search keywords on both your own pages and other sites.

Informational searches handled by AI overviews

It’s also worth noting which types of searches trigger AI overviews. Informational keywords trigger most AIOs (99.2%), while navigational (20.3%), commercial (5.8%), and transactional (4.0%) keywords are less likely to receive AI summaries.

The implication: focusing on mid- and lower-funnel transactional keywords, unique value-add content, and local-focused searches offers more strategic value since these are less likely to be answered with zero clicks by AI.

UGC and community sites heavily weighted

User-generated content and community sites are heavily weighted in AI search.

Community and review sites like Reddit, LinkedIn, and G2 are heavily cited in AI engines, though this has fluctuated over time and by tool/site. Data on the importance of karma/upvotes is mixed: GPT-3 was trained on WebText2 which includes only links from upvoted Reddit posts, but recent Profound research shows karma has limited impact on citations.

Regardless, posts should focus on clarity, relevance, and authority rather than volume.

Structured content and metadata importance

SEO best practices around structured data and image alt-text are even more critical in the AI search era.

Avoid JS-heavy pages which AI may not render or scrape correctly, and use techniques like schema markup, JSON-LD, and FAQs to add structured content that AI can use to contextualize your site.

Evaluating Specific Tactics

The internet is constantly sharing new tactics for AI search. In this section, I’ll walk through a few of these tactics and assess how legit they are (or not).

Dedicated AI/LLM page tactic - not useful

Tactic: Adding a Dedicated Page for AI Agents/LLMs — Likely not useful / misunderstands how web search works. When tested, ChatGPT didn’t reference these dedicated pages at all and used information from other pages deemed more relevant.

llms.txt pages - low impact

Tactic: Adding plain text llms.txt pages — Good for copy-paste if easy to implement, but low impact on AI search. No major AI system currently uses llms.txt, as confirmed by Google’s John Mueller. It’s a proposed standard rather than something actually being used by OpenAI, Google, or Anthropic when crawling websites.

Programmatic content creation with AI

Tactic: Programmatic content creation with AI — Working well for now, but beware algorithm changes. Some companies are seeing significant results (10x ROI on Reddit strategy, +75% citation rate on new pages, 3x increase in AI search citations), but critics point out this is like the spammy, keyword-stuffing era of Google Search. These tactics may work in the short term but could be penalized as algorithms mature.

First-party data importance

First-party data remains the most powerful source of intelligence. Collect self-reported attribution with free-response questions like “How did you hear about us?” to understand key decision influences (categorize after to avoid missed categories and improve authenticity). Additionally, analyze traffic data tagged to AI UTMs like utm_source=chatgpt.com.

Factor 3: Branded Apps and Tooling

Building AI Apps and MCP Servers

Building an AI App or MCP Server can provide your brand with an owned experience within tools like ChatGPT.

Apps offer another potential layer of visibility in AI search with the added benefit of brand control and the ability to facilitate transactions.

The GEO Tool Landscape

Crowded landscape of SEO and GEO tools

Within this context, a crowded landscape of tools are competing for marketing dollars across SEO and GEO.

Traditional players like SemRush and Ahrefs are being joined by new GEO-focused companies including Gumshoe, AirOps, Gauge, Profound, Scrunch, Bluefish, Evertune, AthenaHQ, Otterly.AI, Brandlight, Peec AI, Relixir, Cognizo, Omnia, Brandrank.ai, Goodie, Daydream, Algomizer, Ziptie, RankBee, and Rankshift.

GEO tools primary offering - visibility tracking

For now, the primary offering of AI search / GEO tools is tracking brand visibility over time across models.

These tools show metrics like visibility score, brand industry ranking, share of voice, and average position across different AI platforms.

Where AI Search Tools help most

Where AI Search Tools help most today:

  1. Understanding sources & citations: Understanding which sites are cited for relevant topics can help you prioritize efforts towards building backlinks and mentions for your brand
  2. Gauging brand perception by AI: Evaluating how your brand is perceived by AI today can help guide future positioning and brand work, and address any existing sources of negative sentiment
  3. Identifying and filling content gaps: Charting visibility over time can help identify and prioritize gaps in content and brand presence that can be filled via tactics like UGC, blog posts, or partnerships

While visibility and understanding can be useful, actionability and attribution remain challenging.

Important Caveats on GEO Data

AI search prompts are synthetic

An important caveat: AI search prompts used by these tools are synthetic, and user data is limited and inconsistent.

There’s ongoing debate about the validity of “prompt volume” metrics, for a range of reasons:

Data sharing may increase with AI ads

Data sharing could increase as companies like Google and OpenAI move towards rolling out ads alongside AI tools. Google is now showing ads in AI Mode, and OpenAI announced that ads are coming soon.

Key Voices to Follow

Great follows for GEO and SEO

For ongoing learning in this space, I recommend following:

Always consider incentives when weighing perspectives: Most SEOs want AI search to be viewed as a natural progression of their expertise. GEO companies want this to be viewed as disruptive and urgent. The truth likely lies somewhere in between.


My Key Takeaways

After putting this together, here’s my synthesis:

  1. Don’t panic, but don’t ignore it. AI search is still tiny compared to traditional search, but it’s growing. Treat it as an emerging channel worth monitoring, not an emergency.
  2. Good fundamentals still matter. Quality content, structured data, positive brand mentions across the web—these help with both traditional SEO and AI search.
  3. UGC sites are increasingly important. Reddit, LinkedIn, and review sites carry significant weight in AI systems, both for training data and real-time search.
  4. Be skeptical of silver bullets. Many “GEO tactics” being promoted are either unproven, misunderstand how these systems work, or are short-term arbitrage that may be penalized later.
  5. First-party data is your best friend. Self-reported attribution and UTM tracking will in many ways give you more reliable insights than synthetic prompt data from third-party tools.

The field is evolving rapidly, and much of what we think we know today may change. Focusing on adding genuine value to your users tends to be what lasts when the algorithm changes.

P.S. The original slide deck is also available in Google Slides.