ChatGPT, Google AI, and Perplexity: a local search field guide

Use the product name precisely

“Google AI” can refer to different experiences. Label AI Overviews, AI Mode, and a separate Gemini conversation distinctly in your notes. Similarly, a ChatGPT answer using web search should not be pooled without explanation with a conversation that did not show web sources.

Record the interface label available to you. If a model version or search setting is not visible, write “not shown.” Guessing adds false precision.

What should you inspect in each environment?

Environment	Evidence to capture	Question to ask
ChatGPT	Whether search was used, answer text, cited links, visible product settings	Was the claim supported by a current source?
Google AI Overviews or AI Mode	Exact surface, query, location context, generated answer, source links	Are we measuring this answer separately from nearby maps and organic results?
Perplexity	Answer, source list, visible mode, question and follow-ups	Does the cited page support the statement about the business?

This is an observation guide, not a comparison of unpublished ranking systems. Interfaces and available features can change.

What does platform documentation establish?

OpenAI’s publisher FAQ distinguishes its search crawler, OAI-SearchBot, from GPTBot, which relates to potential model training. A training preference and a search visibility preference are separate decisions.

Perplexity documents PerplexityBot and its published IP ranges. A site owner investigating access should check both crawler policy and network rules.

Google’s AI search guide explains its own search requirements and controls. Do not apply another platform’s bot instructions to Google.

These documents help explain access. Access does not establish that a specific page will be cited.

Use one buyer scenario across products

Choose a genuine customer question with a service and an explicit town. For example: “Which companies repair heat pumps in [town]?” Replace the bracketed text with a real area the business serves.

Start a fresh conversation in each product, record the date and settings, and save the first answer. Do not coach the tool toward your company and then count the result as an independent recommendation.

If you investigate with a follow-up, save it as a separate interaction. Follow-ups are valuable for understanding claims, but they inherit context from the conversation.

Compare meaning, not just name counts

A business might appear in every answer and still be described inaccurately. Another might be absent from recommendations while its educational guide is cited.

Review whether the answer correctly describes the offer, whether any source is obsolete, and whether the proposed next step makes sense for a customer. Note a missing answer or product error separately from a valid answer that does not mention the business.

Preserve the original result when correcting a factual issue. A before-and-after record is more useful than a screenshot taken only after the desired outcome appeared.

What should you change after comparing?

Choose one supported task: clarify an ambiguous service, correct a cited listing, make a useful page accessible, or add evidence to an unsupported claim on your own site.

Then repeat the observation protocol. A later improvement is an observation; multiple site changes and platform changes may prevent a causal conclusion.

Use the measurement method for denominators and repeat runs. For crawler access issues, use the technical SEO guide.