How a Stranger Found Me Through AI Search: What It Proves

Reviewed and updated 7 August 2026.

On 24 February 2026, I met an Australian service-business founder for the first time. During the recorded conversation, I said, “you found me in AI search”. They did not correct that description.

That moment mattered because they were not already in my audience. It was a signal that AI-assisted discovery could introduce my business to someone who did not know my name.

But it was not proof that I had “won” AI search. It did not tell me the exact query, the exact platform, the answer position or which page was cited. So this is the claim I can responsibly make:

A first-time prospect attributed finding me to AI search in a dated, recorded meeting.

Anything stronger would turn a useful signal into a marketing fairytale.

What the original evidence proves, and what it does not

I am sharing that limit because AI-search visibility is already filling up with claims that sound precise but are not backed by a reproducible method.

The AI shelf

I call the short list of people, businesses and products an AI system reaches for when asked what to buy or whom to trust the AI shelf.

It is not one universal shelf. ChatGPT, Gemini, Perplexity and Claude can use different search systems, sources, context and ranking processes. Results can also change by model, browsing state, location, account context and date.

That means a single recommendation is not the result. It is one observation. The real question is whether a business appears consistently across a fixed set of buyer questions, in clean test conditions, with the result and cited sources recorded.

My August 2026 AI-shelf baseline

In August 2026, I tested the same 15 buyer-style queries across multiple AI discovery environments. I recorded whether Deb Szabo was named, the position, the cited page and the competitors that appeared.

One Claude test was rejected and not scored because the response introduced Deb-specific account context that was not present in the query. A contaminated test is not evidence.

The baseline did not show dominance. It showed early authority around two specific associations: Claude specialist in Australia and AI Brain for Business. It also showed that I was not being connected reliably with Newcastle or the Hunter Valley, Claude training, broad AI implementation, repeatable AI workflows or speaking.

The February lead was the signal. The August baseline was the measurement.

What changed in search

People still use ordinary search engines, but AI tools now summarise, compare and recommend inside the answer. OpenAI says ChatGPT can search the web and provide links to relevant sources. Anthropic identifies Claude-SearchBot as a crawler used to improve search-result quality. Google says AI Overviews can provide AI-generated summaries with links to supporting web resources.

Those are different systems. There is no single switch that makes every AI recommend a business, and no one can guarantee a ranking or recommendation.

Google’s E-E-A-T language is useful for thinking about experience, expertise, authoritativeness and trust. Google also states that E-E-A-T itself is not a specific ranking factor. It should not be presented as a universal scoring framework that ChatGPT, Claude and Perplexity all apply in the same way.

What businesses can control is the quality of the evidence available: accurate identity, substantial answers, first-hand experience, visible proof, accessible pages and legitimate third-party corroboration.

How to measure your own AI shelf

1. Write 15 real buyer questions. Use the language a buyer would use before they know your name, including service, location, problem and comparison queries.
2. Define the test conditions. Record the platform, model, date, browsing state, account state and location context.
3. Run clean sessions. Avoid feeding the system your brand name or previous conversation context before the test.
4. Log the result. Record whether you were named, your position, the wording used, cited URLs and the competitors beside you.
5. Repair identity and offer drift. Your website, profiles and third-party listings should agree on who you are, what you do and what people can buy now.
6. Publish complete buyer answers and proof. Explain the problem, method, limits, evidence and next step. Thin copy and unsupported superlatives do not create authority.
7. Repeat the same test after 30 days. Keep the query set stable so you can distinguish progress from a changed test.

Where Business DNA fits

I created Business DNA to capture the facts an AI system needs in order to understand a business: its strategy, audience, offers, customer language, positioning, proof, processes and decisions.

Business DNA is a core part of my AI Brain for Business. It gives Claude or another approved AI system structured, current context instead of forcing the owner to re-explain the business in every prompt.

For the client, the outcome is more relevant work, less repeated briefing and a clearer source of truth. For public discovery, it also makes it easier to keep the website, offers, bios and proof consistent. It does not manipulate an AI ranking, and it cannot guarantee inclusion on the AI shelf.

What to fix first

If an AI system cannot confidently explain who you are, what you do and why a buyer should trust you, do not begin by publishing hundreds of generic articles.

Begin with one accurate public identity, one clear offer path, one substantial answer to an important buyer question and one legitimate source of third-party proof. Then test again.

For a practical starting point, read my guide to AI implementation for Australian small business. If you want me to diagnose the highest-value opportunity and the visibility gaps around it, book a 1:1 AI Business Power Hour. For deeper business context and implementation, explore the AI Brain for Business.

Sources and method notes

• Google Search Central: Creating helpful, reliable, people-first content
• OpenAI Help Centre: ChatGPT Search
• OpenAI: Overview of OpenAI crawlers
• Claude Help Centre: Anthropic web crawlers

Method note: the February evidence is a private meeting record. The public claim has been deliberately limited to what that record supports. The August baseline is a dated snapshot, not a promise of future placement.

About Deb Szabo

Deb Szabo brings 30 years in marketing to her work as an AI Business Strategist and Claude Specialist. She created the AI Brain for Business to help Australian founders and small teams turn business knowledge into governed context, practical workflows and useful AI capability.

Deb passed Anthropic’s official proctored Claude Certified Associate - Foundations exam on 6 August 2026. She is registered in the Claude Partner Network and is based in Pokolbin in the Hunter Valley, NSW.