Skip to main content
Back to Blog
AI Search|7 min read|

How to Tell If AI Search Optimization Is Working

Write down the work, take a starting count before it begins, recheck the same questions on later dates, and count real inquiries on their own.

JD
Justin Dews
Co-Founder, PathOpt

You can tell if AI search optimization is working by tracking three things separately: whether the promised work got done, whether AI tools start naming or linking you on the same questions asked again on later dates, and whether more people contact you. The work needs a written task list before it starts. The other two need a starting count taken before any work begins. One doesn't prove the others, so don't let anyone blend them into one number.

AI search optimization means work meant to get your business named or linked when people ask AI tools like ChatGPT a question. No one can promise it will.

PathOpt is holding its own business to this test right now.

Key Takeaways

  • Before work begins, get a starting count: fixed questions, named AI tools, a date, and each tool reported on its own.
  • Expect answers to swing. Recheck before you count a change. One good day can be luck.
  • We haven't shown a lift yet. We've published our starting count and will publish the recheck, even if nothing changes.
  • What are the three things to track?

    Whatever a provider calls the service, hold the work to three measures. Keep each one separate, because each can move without the others.

    What to track What it tells you How you check it
    1. The work is done Did they do what they said they'd do? Each task on the written list is either finished or it isn't.
    2. AI tools name or link you Are the answers changing? Ask the same questions on the same tools on later dates and compare with the starting count.
    3. More people contact you Is it reaching your business? Count real calls, forms and requests against the count before work began.

    The work can be finished while the answers stay the same. The answers can change while the phone stays quiet.

    Our How We Measure page explains how we count leads, and What Your Agency Should Be Showing You covers the reports to expect.

    What does a fair starting count look like?

    A fair starting count has four parts: a fixed list of questions, named AI tools, the date the questions were asked, and results for each tool reported separately.

    Here's PathOpt's starting count. On Oct 27, 2026, PathOpt asked six AI tools 24 questions a small business owner might ask, plus 2 that name PathOpt, 3 times each. The tools were ChatGPT, Google Gemini, Google AI Mode (Google's AI answers for harder questions), Google AI Overviews (the AI summary at the top of Google results), Claude and Perplexity. For Claude and Perplexity we used the versions other software uses, not the apps.

  • Customer questions: PathOpt was mentioned or linked in 0 of 430 answers.
  • Questions naming PathOpt: 36 of 36 answers found it.
  • The full results of PathOpt's AI search test list every tool separately and explain how we asked. Those are the numbers we'll be held to. The first recheck is planned about 4 weeks after our new pages show up in Google, then on a few more dates after that, with the same 26 questions on the same 6 tools.

    Why do the results swing from one check to the next?

    AI answers change, so a single check is a snapshot, not a verdict. Every answer in PathOpt's starting count came from one evening, and the 3 tries were minutes apart. That tells us nothing yet about change across days or weeks.

    Wording matters too: in our Nov 3, 2026, test checks of three Dallas-area home service businesses, one was named in 3 of 3 answers to one phrasing of a hiring question and 0 of 3 to another, each asked once on three AI tools, as our guide to checking AI mentions shows.

    Google's own guide says AI Overviews and AI Mode "may use different models and techniques, so the set of responses and links they show will vary" (Google Search Central). So check each tool on its own.

    Recheck before you count a change. One good day can be luck.

    What does Google give you to measure with?

    Google gives you one report built for this. Search Console, Google's tool that shows how your site does in Google search, has a Generative AI performance report. Google rolled it out to all sites on Aug 31, 2026.

    The report shows impressions, meaning how many times links to your site were shown in AI Overviews and AI Mode. You can break that down by page, country, date and device (Google Search Console Help).

    It doesn't show the questions people asked, the answers they got, times your business was named without a link, or who else was named. A site with too few impressions may not see the report at all. We haven't looked at our own report yet.

    So ask your provider for this report as one input, not the whole answer. That gap is why someone still has to ask the AI tools your questions and save what they say.

    The same Google guide adds that a page must be indexed, meaning Google has added it to its search results, before it can appear as a link in AI Overviews or AI Mode. So checking that new pages are indexed is a fair early step.

    Also ask your provider to confirm that your Search Console setting under Settings, then Search generative AI, includes your site in Google's AI features.

    What should I ask my AI search provider?

    Ask these before you pay and at each report.

  • What questions are you tracking, and can I see the list? A good answer is a written list that stays the same between checks.
  • Which AI tools do you check? A good answer names each tool and reports each one separately.
  • What was my starting count, and when was it taken? A good answer is a number for each tool, with a date, from before work began.
  • Can I see the saved answers? A good answer is yes, with the actual answers and links, not just a summary.
  • When will you check again? A good answer gives set dates, using the same questions on the same tools.
  • How do you count real inquiries, and what was my count before work began? A good answer is a count of calls, forms or requests, kept apart from the AI results.
  • What work will you do, and how will I know it's done? A good answer is a written list of tasks, each one marked finished or not.
  • What can't your check tell me? A good answer states limits, such as one day's answers, and admits what they can't see.
  • What are the red flags?

  • A guarantee. Nobody can promise that an AI tool will name or recommend your business. Our own search engine optimization page says inclusion in AI answers can't be guaranteed.
  • A single blended score. One number hides which of the three measures moved, if any did.
  • No saved answers. If you can't see what the tools said, you can't check the claim.
  • Changed questions passed off as a fair comparison. New questions can't be compared with old ones as a before and after. Only the questions asked both times can.
  • To judge an audit itself, use our checklist, What a Small-Business AI Visibility Audit Should Include.

    Has PathOpt shown that this works?

    We haven't shown it yet. We have a starting count, not a result.

    Some of the work is done. A file on our site written for AI tools had a wrong line about how we charge, and we fixed it on Nov 3, 2026. We're writing pages that answer the questions owners ask. This article is one of them.

    When the recheck is done, we'll publish it on the test results page, next to the 0 of 430 and 36 of 36. If nothing changes, we'll say so plainly.

    What this can and cannot tell you

    These measures can't show why an AI tool picked one source over another. We can see what the tools chose, not their reasons. They also can't show how often customers ask your questions, and any single check reflects one day's answers.

    Facts on this page last checked Nov 10, 2026.

    Want a free first snapshot? PathOpt's free AI visibility audit turns the services and cities on your website into questions customers might ask, adds up to 3 real questions people asked ChatGPT, and asks each one once on ChatGPT, Google AI Overviews, Google AI Mode and Perplexity. It's a one-day snapshot that totals each question across the four tools, not a starting count for each tool.

    The report lists every question we asked and how often you came up for each one, the businesses and sites that showed up instead, what we found on your Google profile, and one thing to fix first. It doesn't include the full text of each answer. A person reads every report before it goes out, and a call is offered, never required.

    Request a free AI visibility audit

    JD
    About the Author

    Justin Dews

    Co-Founder, PathOpt

    Justin Dews is a founding partner at PathOpt, where he helps small businesses grow through performance marketing, automation, and operational systems. He writes about running marketing accountably: full account ownership, transparent reporting, and spend tied to real revenue.

    Warm gradient in PathOpt's brand colors, from terracotta to cream
    Ready to Stop Drowning?

    Your Marketing Works. Can Your Business Handle It?

    Get the strategic thinking, performance marketing (Google Ads, Meta, SEO), AND the AI automation to actually capture, convert, and handle the business it brings—from three partners who’ve built this themselves.