JustHandled Labs
Menu
FIELDNOTES / BUSINESS DISCOVERY

Can AI search help someone choose your business?

A mention is a start. An accurate answer and a useful landing page give it somewhere to go.

Conceptual AI-generated illustration: a ceramic shop owner welcomes a visitor carrying a phone. A metaphor for discovery, not an actual customer or measured result.
Conceptual AI-generated illustration. Not an actual customer, reporting interface or measured business result.

Someone asks an AI assistant where to take a beginner pottery class. Your studio might appear in the answer. So might an old listing with last year's dates. Either way, a mention does very little for a customer who still cannot work out what to book.

That is the useful starting point for AI search: can people find an accurate answer about your business, then reach a page that helps them decide? Before buying another dashboard, look for the gap between those two steps. The reports you already have can give you somewhere sensible to start.

Four services, four different windows

There is no single scoreboard here. Google can show AI impressions, Bing can show citations, and your own analytics can show some visitors arriving from ChatGPT or Perplexity. Each tells you something useful. None tells the whole story.

Reporting available in public documentation reviewed October 9, 2026
ServiceStart hereWhat the signal cannot tell you
Google SearchSearch Console’s Generative AI performance report: impressions and pages in AI Overviews and AI Mode, with country, device and date views.This report describes visibility, not a separate AI conversion funnel. Use your site analytics for visitor actions.
Bing / CopilotBing Webmaster Tools’ AI Performance: citations, cited pages and sampled grounding queries; preview features also include intents, topics, citation share and comparisons.Supported Microsoft and partner surfaces are aggregated. Citation share is observational, not a universal rank or market-share measure.
ChatGPTYour analytics: OpenAI says search referrals include utm_source=chatgpt.com. Inspect landing pages and actions for that source.Referral visits do not reveal every answer that mentioned or cited you. The publisher FAQ does not document a Bing-style sitewide citation report.
PerplexityYour existing analytics for identifiable referrals, plus a small dated sample of questions and their cited links.Do not assume an all-site citation dashboard. Perplexity describes analytics for Publisher Programme partners; that is not equivalent to general webmaster reporting.

Google's AI report announcement carries an August 31 worldwide rollout note. Its report guide explains the views; low impressions can affect availability, and Search Labs experiments are excluded. AI activity also appears in overall Search performance, so adding those totals would count it twice.

Bing's AI Performance report and expanded preview help you spot cited pages and topics worth clarifying. For ChatGPT, use the referral tracking described in OpenAI's publisher FAQ. Perplexity's partner analytics are a different offering; an ordinary business should start with identifiable referrals in its existing tools.

If crawler access is the issue, check the service's own instructions: OpenAI distinguishes OAI-SearchBot from training access, and Perplexity's crawler guide distinguishes search discovery from user-requested fetches. Allowing a crawler does not promise a citation.

Give yourself something useful to compare

Pick three public pages tied to customer decisions: your main offer, a comparison or “who this is for” page, and one useful how-to. Choose one meaningful action for each, such as a qualified enquiry, a completed booking, or a resource download. A click toward checkout is an intermediate action; a paid order needs its own confirmation.

  1. Save the most recent complete 28-day reports you can access, with their dates and filters. Record identifiable AI referrals and the actions visitors took.
  2. Write down five realistic customer questions. Try them once in each service you can access; save the question, date, relevant location or language, search mode, cited link and any incorrect detail.
  3. Choose one page improvement and record when you made it. Compare another complete period, noting other campaigns, offer changes and seasonal effects.

No analytics yet? Start with the available report exports and ask enquirers, “How did you find us?” Keep their answers separate from measured referrals. Neither needs to become a complicated reporting project before you can fix an obviously confusing page.

The class dates are buried. Fix that first.

Synthetic example: the workshop and all numbers below are invented to demonstrate a decision. These are not JustHandled results.

A ceramics studio wants more suitable beginner-workshop bookings. During one complete 28-day period, Google shows 180 AI impressions for its workshop page; Bing reports 14 citations; its analytics identifies 9 ChatGPT referral visits and 3 Perplexity referral visits. Two of those ChatGPT visitors request a date. The owner does not add impressions, citations and visits into “206 leads.” They measure different things.

In this hypothetical question sample, two answers give an outdated class length. On the studio's own page, dates sit below a long story about the joys of clay. The story may be lovely. A beginner trying to plan Saturday needs the dates.

The owner corrects the duration and materials, moves upcoming dates and the booking step beside the answer, and adds a current photo with a clear caption. Next month they compare enquiries and bookings alongside visibility. If citations rise while suitable enquiries stay flat, the next job is to examine the audience and the booking path. Writing five more articles would be a fairly expensive way to avoid looking at the page.

Match your improvement to the gap

  • Your page cannot be reached: check public access, indexing eligibility, canonical URLs, crawler rules and internal links. Our free robots and sitemap guide helps diagnose conflicting discovery signals. It is a technical starting point, not a promise of inclusion.
  • Answers misdescribe your work: state the service, intended customer, scope, current price or pricing approach, location and limitations plainly. Update conflicting details on your own public pages and authorized business listings.
  • People find you but cannot decide: add a direct answer, a concrete current example, useful comparisons and supporting sources. Put the key information in readable page text, not only inside an image.
  • Visitors arrive but do not act: check the landing page on a narrow phone screen. Make the relevant next step clear, with prerequisites and price before a purchase commitment. Improve the offer or navigation before chasing more mentions.

Google’s AI-feature guidance emphasizes accessible content, internal links and structured data that matches the visible page; it does not require special AI markup. Bing recommends clear structure and evidence. These practices make pages easier to use and interpret, but none guarantees a citation.

Start with the question your customer actually asks

Pick one page and one question: “Is this class suitable for a complete beginner?” “Do you deliver here?” “What will this cost?” Check the answer, then follow its link on your phone. Can you make the decision without a scavenger hunt?

Save the available reports, fix the most useful gap, and set a date to compare the next complete period. You do not need a paid monitoring tool to do that. You need a page that earns the visit.

A few practical notes

Compare matching date ranges and note time zones; Google's report uses Pacific Time. Unavailable reports and unattributed visits are gaps in knowledge, not proof that nobody saw you. Keep your own checks out of visitor totals where possible, and keep customer information out of test prompts. A handful of sampled answers can uncover mistakes; it cannot establish a universal ranking or prove what caused a booking.

By JustHandled Labs. Based on public first-party documentation reviewed October 5, 2026. The studio and its numbers are hypothetical, not customer results or firsthand testing. Reporting features can change.