Who does AI recommend? What the evidence shows across 4 AI engines

What the four engines actually do

ChatGPT Search can search the web and return answers with inline citations or a sources panel. Google’s AI Overviews and AI Mode are designed to help users explore information while linking to supporting pages. Perplexity likewise presents answers with cited sources. These systems are not identical, and none publishes a complete formula for how local businesses are selected.

• ChatGPT Search: uses web search when current information is needed and may show inline citations and additional sources.
• Google AI Overviews and AI Mode: synthesise information within Google Search and provide links that let users continue to relevant sites.
• Perplexity: is built around answer generation with cited web sources, although the exact source mix can change between prompts and runs.
• Gemini: can draw on Google’s information ecosystem, but a Gemini response and a Google AI Overview should not be treated as the same product or the same result set.

What published GEO research supports

The foundational Generative Engine Optimisation study introduced a framework for measuring visibility inside generative answers. In its controlled benchmark, some content changes improved visibility by as much as 40%. That result is important, but it should not be misrepresented: it was produced inside a defined experimental setting and does not prove that any tactic guarantees organic discovery, local recommendations or lasting traffic across commercial AI products.

A 2026 critical survey of GEO research reached a similarly cautious conclusion. It described AI visibility as a multi-stage and partly stochastic process involving discovery, retrieval, reranking, citation and user behaviour. The review found that relevance and context are among the more reproducible factors, while broad “one-size-fits-all” optimisation rules transfer poorly between platforms and domains.

What can reasonably influence a local recommendation

No single factor has been proven to control recommendations. However, the following practices improve the quality, accessibility and corroboration of the information an AI system may retrieve:

• Accurate entity information: keep the business name, location, services, contact details and opening information consistent across the website and trusted profiles.
• Direct, verifiable answers: publish pages that clearly explain who the service is for, where it is available, what it includes, limitations, pricing approach and next steps.
• Crawlable technical foundations: use stable canonical URLs, descriptive titles, internal links, structured data where appropriate, and avoid blocking important pages from crawlers.
• Independent corroboration: credible directories, associations, editorial coverage, partner references and customer review platforms can help confirm that a business exists and is relevant.
• Freshness where freshness matters: update time-sensitive service, location, pricing and availability information instead of relying on stale pages.

Why engines may recommend different businesses

Different answers are expected because engines can use different indexes, retrieval systems, model versions, freshness windows and ranking signals. The wording of a prompt also changes intent. “Best”, “most affordable”, “near me”, “for first-home buyers” and “open on Sunday” are different requests and may legitimately produce different recommendations.

Results can also vary between repeated runs. For that reason, a single screenshot is weak evidence. A credible audit should repeat prompts, record dates and locations, separate citations from unlinked mentions, and distinguish brand visibility from actual referral traffic or enquiries.

A defensible way to measure AI visibility

1. Define a fixed set of real customer prompts by service, location and buying stage.
2. Test each prompt across the relevant engines using a documented date, location and account state.
3. Repeat each prompt and include natural paraphrases rather than relying on one wording.
4. Record whether the brand is named, recommended, cited, linked or absent.
5. Capture the cited domains and pages, not just the generated answer.
6. Re-test after meaningful website or authority changes and compare like with like.
7. Keep AI visibility separate from commercial outcomes such as qualified visits, calls and sales.

What Australian local businesses should do now

Start with accuracy and evidence. Make every service and location page useful to a real customer, keep business details consistent, publish original expertise, and earn independent references that can corroborate important claims. Then measure visibility across more than one engine over time.

Avoid claims such as “schema makes AI recommend you” or “a certain number of reviews guarantees inclusion”. Those statements are not supported by the available evidence. Structured data, reviews and third-party mentions may be useful parts of a broader information ecosystem, but none is a universal switch.

Sources and limitations

This article is based on public documentation and research available on 30 July 2026. Commercial AI systems change frequently, and their complete ranking and recommendation systems are not public. The article therefore describes supported observations and practical implications, not guaranteed ranking factors.

• OpenAI Help Center: ChatGPT Search
• OpenAI: Introducing ChatGPT Search
• Google: How AI Mode and AI Overviews help people explore the web
• Aggarwal et al.: GEO — Generative Engine Optimisation
• Martinez: Critical survey of GEO research, 2023–2026