How to win in AI search, and put AI to work in your business

What is GEO?

GEO, or generative engine optimization, is the work of getting your content surfaced and cited inside AI answer engines like ChatGPT, Perplexity, Google AI Overviews and AI Mode, Gemini, and Claude.

The term comes from a 2024 research paper by Aggarwal and colleagues, presented at KDD 2024, which showed that specific content tactics can raise a source’s visibility in AI answers by up to 40 percent across a benchmark of 10,000 queries (Aggarwal et al., GEO, KDD 2024). You will also hear it called AEO (answer engine optimization) or LLMO (large language model optimization). They describe the same goal: be the answer, not just a result.

How is GEO different from SEO?

SEO optimizes a whole page to rank in a list of links. GEO optimizes for being extracted and cited inside a single AI generated answer, often at the passage level rather than the page level.

Mike King of iPullRank calls this relevance engineering: AI search competes at the passage level, pulling fragments of pages rather than whole documents, and a passage can be cited almost independently of where the page ranks (iPullRank, How AI Mode Works). The other practical difference is off site. GEO success leans heavily on brand mentions across the web, not just your own pages. The overlap with SEO is still large. As Lily Ray of Amsive puts it, good SEO and good GEO are mostly the same work, because AI engines lean on the same index and authority signals Google already uses (Amsive).

Why does AI search matter now?

Because the click is disappearing from the top of search, and AI answers are taking its place.

SparkToro’s 2026 study found that fewer than one third of United States Google searches still send a click to the open web, with roughly 68 percent ending without one (SparkToro, 2026). Google AI Overviews now appear on more than a fifth of searches and reduce click through when they show. The takeaway is not that your website stops mattering. It is that raw traffic is no longer the only scoreboard. When an AI answer names you as the source, that mention does the early selling, and AI referrals tend to convert better on high intent pages because the reader arrives pre qualified.

What actually gets you cited by AI?

Two things drive most of it: strong classic search authority, and a wide brand presence across the web. The on page tactics that measurably help are citing statistics, adding expert quotations, and citing your sources.

The single most consistent finding across the 2025 to 2026 research is that AI visibility is largely downstream of work you already understand. Lily Ray studied a set of sites through a 2026 Google update and found organic traffic and AI citations fell together, concluding that the fastest way to lose visibility in AI search is to lose it in Google first (Amsive). Ahrefs analyzed 75,000 brands and found brand mentions, not backlinks or page count, were the strongest correlate of AI visibility, with YouTube mentions the single strongest factor (Ahrefs, 2025).

Lift figures from Aggarwal et al., KDD 2024, measured on a 10,000 query benchmark. One honest caveat the paper notes: citing sources can lower visibility for an already top ranked page, so it lifts underdogs most.

What is hype versus what actually works?

The biggest gains come from authority and brand presence. The most oversold ideas are AI specific files and tricks that promise to bypass weak content.

On schema and llms.txt, be clear eyed. Google states there is no special structured data or text file you need to appear in AI features. Keep your normal schema accurate because it helps machines parse you, and treat llms.txt as a cheap, low risk experiment rather than a priority. The work that compounds is authority, mentions, and genuinely useful, well sourced content.

Which AI crawlers should you allow?

If you want to appear in AI search answers, allow the search and indexing bots. Training bots are a separate decision, and the names are easy to confuse.

Two points clients get wrong. Blocking Google-Extended does not remove you from Google AI Overviews, because Overviews use the regular Google index. And blocking GPTBot is a training decision that is separate from your ChatGPT search visibility, which is governed by OAI-SearchBot (Google Search Central).

How do you measure AI search visibility?

Track three layers: AI referral traffic, AI bot crawl activity, and your citation share of voice across engines.

AI sends little referral traffic today, but it is growing fast and converts well on high intent pages, so the scoreboard shifts from clicks to mentions. Build a custom channel in GA4 that catches referrers like chatgpt.com, perplexity.ai, and gemini.google.com, knowing much AI traffic arrives with no referrer and lands in Direct. Check server or Cloudflare logs for the AI bots above. Then track citations and share of voice with a tool such as Profound, Otterly, or Peec AI. One caution from Aleyda Solis and others: AI answers are inconsistent, so test 30 to 50 commercially relevant prompts across two or three platforms on a documented rubric, and do not blend the platforms into one number (Aleyda Solis, AI Search Optimization Checklist).

Where should you start with AI in your business?

Pick one repetitive, high volume, low judgment task, pilot it with a person reviewing the output, and measure the time you save against a baseline you capture first.

Do not try to transform everything at once. A simple loop works: identify the repetitive tasks, pick one, pilot it with a human in the loop using a business tier tool that does not train on your data, write a one page usage policy, measure time saved, then expand to the next use case only after the first proves out. This maps onto the NIST AI Risk Management Framework (govern, map, measure, manage), so you can say you follow a recognized standard (NIST).

Crawl, walk, run

• Crawl. Individuals use assistants for drafting and summarizing, with review before anything ships.
• Walk. One or two workflows are standardized, governed by a short policy, and measured for time saved.
• Run. AI is grounded in your own data, automations connect across your tools, and you track it against business metrics.

What is the honest state of AI adoption?

Almost everyone is using AI, but few are capturing real value, and the gap is people and process, not technology.

McKinsey’s 2025 State of AI found about 88 percent of organizations use AI in at least one function, yet only around 6 percent attribute meaningful value to it (McKinsey). BCG’s read is the most useful framing for an owner: 74 percent of companies struggle to scale value, and success follows a 10 20 70 rule, roughly 10 percent algorithms, 20 percent tech and data, and 70 percent people and process change (BCG). For small businesses specifically, a 2025 survey found 76 percent are using or exploring AI, with marketing and customer engagement the top impact area (Reimagine Main Street).

Which AI use cases pay off first?

The fastest wins are in customer service, content and marketing, sales operations, and turning your own data into plain language answers.

How do you use AI safely with customer data?

Use a business tier tool that does not train on your data by default, keep a person reviewing anything customer facing, and never paste confidential data into a consumer chatbot.

The key distinction to teach your team: consumer tiers of the major assistants may use chats to improve models, while business and enterprise tiers and APIs do not train on your data by default. Note the nuance, not trained on is not the same as not stored, since business data still lives on the vendor’s servers under their retention policy. Control hallucinations by grounding answers in your trusted documents and requiring citations, and document your brand voice so output sounds like you. For claims about what your AI does, the Federal Trade Commission is blunt: do not exaggerate, and keep evidence for performance claims (FTC).

What is the ROI of AI, and how fast?

Task level gains are real and well documented, and they help less experienced staff the most. Expect time savings in 30 to 60 days and deeper business impact in 3 to 6 months.

A meta analysis by Nielsen Norman Group found an average productivity uplift around 66 percent across studied tasks (NN/g). A study of 5,179 customer support agents found 14 percent more issues resolved per hour, and 34 percent for the newest staff (Brynjolfsson, Li, Raymond, NBER 2023). The reason most companies still fail to capture value is not the tools, it is the missing baseline and process. So capture a before number, measure time saved and conversion lift, and expand only what proves out.