On September 17, 2026, Search Engine Roundtable’s Barry Schwartz reported that he could replicate a change first posted by local search consultant Ben Fisher: Google’s mobile local knowledge panel — the card your Google Business Profile populates — now carries an “AI Overview” label and a “Show more” button that expands into an AI Mode chat interface. The paragraph describing your business there is no longer a field you filled out. It is written for you, from whatever Google can read about your company.
For a single-location shop that is a curiosity. For an HVAC company running eleven profiles across Warren, Sterling Heights, Livonia and Rochester Hills, it is a content-freshness problem that just became customer-facing.
What exactly changed in the Google local knowledge panel?
The business description block inside the mobile knowledge panel now renders as an AI Overview rather than as your written profile text. Schwartz documented two elements: an “AI Overview” label above the summary, and a “Show more” control that drops the searcher into the same follow-up chat surface Google uses for AI Mode.
The second element matters more. A labeled summary is a display change. A button that opens a conversation means the next question — pricing, commercial accounts, Sunday hours — gets answered by a model reading your public footprint, before anyone reaches your phone number. Google has circled this for a year with AI review summaries and AI-written profile descriptions, but this is the first sighting where the panel itself is labeled an AI Overview.
Where does Google get the text for the AI summary about my business?
From your public content — website copy first, then reviews, then your profile fields. Schwartz gave away the mechanism when he noted the summary described his firm’s “niche solutions” using outdated information and blamed his own corporate site rather than the model: the copy was stale, so the summary was stale.
That sentence tells you how to predict the output. The old panel was a data surface: you typed a description, picked categories, set hours, and Google displayed them. Accuracy was a data-entry problem. The new panel is a synthesis surface: Google reads what is crawlable about the entity and compresses it. Accuracy is now a content-maintenance problem across every page that mentions you.
Run that forward. A Southfield law firm that dropped its immigration practice in 2023 but still has a 2019 practice-areas page live gets a panel summary listing immigration law. Nothing in the Business Profile is wrong. The profile is not the source anymore.
Is this confirmed, or is it still a test?
Unconfirmed. Google has published no announcement, Help Center entry, or changelog. What exists is Ben Fisher’s LinkedIn post and Schwartz’s replication on mobile — two credible sightings, not a launch. So rebuild nothing on the assumption it ships as seen. The playbook below pays off either way, because the same stale pages that produce a bad panel summary already produce bad AI Overviews and bad ChatGPT answers about your business today.
Does an AI summary actually cost a business clicks?
The closest evidence is Ahrefs’ France study, published August 19, 2026. Google launched AI Overviews and AI Mode in France on July 22, 2026, and Ahrefs measured 963 French domains across the 28 days before launch against the 9 days after. Domains where more than 20% of queries triggered an AI summary lost 23.1% of click-through rate on average. At the median across all 963 domains, the decline was 5.7% of CTR and 8.8% of clicks.
The gap between 23.1% and 5.7% is why “AI Overviews cost sites 23% of their clicks” misreads it. The larger figure belongs to heavily exposed domains; exposure is the variable, not AI generally. Ahrefs found healthcare took the worst combination — 22.2% AI presence, a 20.4% CTR drop — the profile of a business whose queries are informational and answerable in a paragraph.
Applied to the panel, that gets specific. “What does this company do” is fully answerable, so the summary absorbs the click. “Can you be here Thursday to look at a rooftop unit” is not. Businesses that sell an explanation are exposed; businesses that sell an appointment are largely insulated — provided the summary describes them well enough that the searcher still wants the appointment.
Here is where I would push back on Google. Web AI Overviews have levers: nosnippet, data-nosnippet, max-snippet. The knowledge panel has none. An owner who does not want a model paraphrasing their company has no toggle to reach for — and Schwartz said plainly he was not sure he wanted AI describing his business when he could describe it himself. When a change removes control rather than adding capability, the burden of proof sits with the vendor. It has not been offered.
Why does this hit metro Detroit multi-location brands hardest?
Because multi-location service companies here almost universally run near-duplicate location pages backed by one real services page, and that architecture inverts badly under synthesis. The pattern is familiar from any Michigan home-services or dental group: eleven location pages for Royal Oak, Troy, Dearborn and Livonia, each a swapped city name over identical body copy, all pointing back to a corporate services page last rewritten three or four years ago. Under the old panel, errors stayed local to one profile. Under synthesis, every location inherits the same corporate source — one stale paragraph propagates across all eleven panels at once.
The stakes are concrete. A Livonia dental group that stopped taking pediatric patients in 2024 still has its 2019 services page live. The panel tells a parent the practice sees children. They call, the front desk says no, they book elsewhere — and some fraction leave a two-star review about being misled. You lost the patient and a rating point on an impression you never paid for.
Seasonality sharpens the timing. Metro Detroit heating calls ramp with the first sustained sub-40°F nights in late October, and a furnace emergency is an unforgiving moment for a summary that gets your service area wrong. If you are reviewing how your local search presence is structured before Q4, this is the season.
How should a small team audit this before Q4?
Seven steps, doable by a three-person team in a week. No enterprise tooling.
- Pull the summary you are actually getting. On mobile, search your brand plus city and screenshot the panel — every location, not just the flagship. Repeat logged out; personalization changes what you see.
- Ask the panel the follow-up questions. Tap “Show more” if you have the treatment, then ask what a customer asks: services, pricing, hours, service area. Write down every wrong answer.
- Trace each wrong answer to a page. Search
site:yourdomain.complus the wrong term. The source is nearly always a page you forgot existed — an old service page, a 2021 press release, a location page for an address you left. - Fix or remove the source, do not just add a correction. A new page saying “we no longer offer X” while the old page still says you do gives the model two sources and no tiebreaker. Update the original URL or 410 it.
- Give each location page real, different content. Named technicians, service radius by township, brands that location stocks, hours that differ from corporate. Duplicate pages give synthesis nothing location-specific, so it falls back to corporate copy.
- Tighten your Business Profile fields anyway. Categories, services, attributes and hours remain the highest-confidence signals Google holds about you. They constrain the summary even when they no longer write it.
- Set a 30-day recheck. Summaries regenerate; a fix that lands in October can drift in December when a new page publishes.
Step three is the one teams skip and the one that returns most. Few businesses keep an inventory of what their own domain says about them, which is why the same outdated claims keep surfacing across Google, ChatGPT and Perplexity. An audit of the pages feeding those answers usually finds them in an afternoon.
What dates should Michigan marketers have on the calendar?
Two verified platform dates and two seasonal ones — the seasonal items are Michigan demand patterns, not Google deadlines.
- August 19, 2026 — Ahrefs published the France CTR study. Read it before accepting any “-23%” claim secondhand.
- September 17, 2026 — the panel treatment was independently replicated. Date your screenshots from here so you have a before-and-after if it rolls out wider.
- Late October 2026 — metro Detroit heating demand ramps. Finish panel accuracy work for HVAC and plumbing before this, not during.
- November–December 2026 — holiday hours and closures. Every hours exception you fail to post is a sentence the summary gets wrong for someone standing outside your door.
Frequently asked questions
Can I opt out of the AI Overview in my knowledge panel?
No opt-out exists as of September 18, 2026. The nosnippet and data-nosnippet directives govern what Google quotes from your pages and may limit what the summary draws on, but no Business Profile setting disables the treatment. Google has not announced the feature, so no controls have been documented with it.
Will rewriting my Business Profile description fix a wrong summary?
Probably not on its own. Schwartz’s replication points to website content as the dominant input — his profile was not the problem, his corporate site was. Rewrite the description because it still feeds other surfaces, but expect the real fix to come from correcting or removing the page carrying the stale claim.
Does this affect desktop searches too?
The sightings so far are on mobile knowledge panels. Google has historically tested local surfaces on mobile first and extended to desktop later, but that is a pattern, not a commitment. For most Michigan service businesses mobile is where near-me demand lands anyway, so mobile-only is no reason to wait on the audit.
Should a single-location business worry about this?
Less than a multi-location brand, but the same audit applies at a fraction of the effort. A shop with a five-page site can screenshot its panel, ask three follow-up questions and check every page against the answers in under an hour. Exposure is smaller because there are fewer stale pages to contradict you.
Comments are closed