Because AI citation is not Google rank. Microsoft Copilot builds its answers from Bing's index and from pages whose answers are written out in quotable text — and in Michael Putnam's second field study, his own Sarasota County home-value site was cited 205 times by Copilot in 103 days while Google ranked it at average position 30.2. Read alongside his first study — in which 70% of page-one Google Maps businesses were never named by ChatGPT — the two results are the same finding from opposite directions: the list AI engines assemble does not follow Google's order, in either direction.
The study: "How AI Answers Cited a Sarasota County Home Value Website: A 1-Site Case Study, July to October 2026," data collected and published by Michael Putnam, a licensed real estate agent with Putnam Realty Group in Nokomis, Florida, who builds and maintains the site it measures, Florida Home Value AI (floridahomevalueai.com): https://floridahomevalueai.com/ai-visibility-study-2026 — data by Michael Putnam; republished free with the author's permission and full credit, under the republication terms stated in the study itself; this write-up and analysis by ScriptMasterLabs. It is the second study of his we have republished; the first, run under his PresenceScanner name, found that 70% of top local businesses are invisible to AI.
The quotable measurement: In Michael Putnam's 2026 case study, one Sarasota County home-value website was cited 205 times by Microsoft Copilot between 14 July and 24 October 2026 — its citations per day growing from 0.8 in July to 3.2 in October — while Google showed its pages about 5,260 times at an average position of 30.2 and sent it 72 clicks. Google's own AI features showed the site 702 times over a longer window. And when ChatGPT was asked which local website shows home values in one of its communities, it listed the site first, above Realtor.com.
How the study was run
The numbers come from five readings, the first four read on 26 October 2026. No paid ads, no boosted posts and no bought links were involved; besides search and AI answers, the site's only promotion was free posts in Facebook groups and letters mailed to homeowners.
| Source | What it reports | Dates covered |
|---|---|---|
| Bing Webmaster Tools, AI Performance (beta) | Times Microsoft Copilot and its partners cited a page | 14 Jul to 24 Oct 2026 |
| Google Search Console, Performance (Web) | Clicks, impressions, average position, searches | 25 Jul to 24 Oct 2026 |
| Google Search Console, Generative AI features (beta) | Times the site's pages were shown in Google's AI features | 31 May to 24 Oct 2026 |
| ChatGPT, signed out | 1 question, asked once that day, answer saved | 26 Oct 2026 |
| ChatGPT, earlier test | The same kind of question for 4 communities, from his notes at the time (whether he was signed in wasn't recorded) | 19 Aug 2026 |
Finding 1: Copilot cited the site 205 times, and the daily pace grew
| Month (2026) | Citations per day | Total citations |
|---|---|---|
| July (from the 14th) | 0.8 | 14 |
| August | 1.4 | 42 |
| September | 2.4 | 72 |
| October (to the 24th) | 3.2 | 77 |
July and October are partial months, which is why the study reports citations per day. The rise was not steady: September started slower than August, with 6 citations in its first 10 days, then had 66 in its last 20. A single AI answer can cite more than one of the site's pages, so 205 citations does not mean 205 answers. Nineteen pages were cited in all; the leaders were written-answer pages:
| Page | Copilot citations | What it answers |
|---|---|---|
| Palmer Ranch guide | 48 | What Palmer Ranch is, what homes cost, its communities, schools and the Legacy Trail |
| Flat fee MLS | 31 | What a flat fee listing is and what each service level costs |
| Homestead exemption | 31 | How Florida homestead works and how to file |
| Talon Preserve community site | 16 | What homes in Talon Preserve sold for, by home type |
| Gated communities | 14 | HOA vs CDD fees, and what gated communities in Sarasota County cost |
| IslandWalk community site | 12 | What homes in IslandWalk sold for, by home type |
Thirteen more pages were cited 1 to 9 times each. Neither of the site's two calculator pages was among the cited pages Bing listed — a point Finding 4 and the pattern section below return to.
Finding 2: On Google, the same site's average position was 30.2
| Google Search Console, 25 Jul to 24 Oct 2026 | Figure |
|---|---|
| Clicks | 72 |
| Impressions | about 5,260 |
| Click rate | 1.4% |
| Average position | 30.2 |
| Searches Google listed | 473 |
Broad searches showed the site but sent nobody — "venice median home price" drew 98 impressions and 0 clicks. Google hides the search behind most clicks when too few people typed it; only 2 of the 72 clicks can be tied to a search. Putnam also notes that some of the 473 listed "searches" are not searches at all ("yes", "both", "howd you know") — consistent with reporting that the general Search Console report includes AI Overviews and AI Mode, and that a follow-up question in AI Mode counts as a new search. Some of the 72 clicks and about 5,260 impressions may therefore have come through Google's own AI answers; Search Console does not separate them in this report, so that cannot be checked.
Finding 3: ChatGPT listed the site first, above Realtor.com
On 26 October 2026, signed out of ChatGPT, Putnam asked: "What local website shows home values in Talon Preserve, Nokomis FL?" ChatGPT listed the site's Talon Preserve page first and Realtor.com's property records second, closing with: "Start with the first website. It is specifically designed for Talon Preserve..."
On 19 August 2026, a similar question was asked for four communities. ChatGPT listed the site first for Talon Preserve, IslandWalk and Gran Paradiso, and did not name it for Palmero — three of four. Those August results come from his notes at the time. One more detail matters for anyone treating AI answers as current: ChatGPT quoted medians the page published in August, from sales through 3 August 2026. By 26 October the page covered sales through 2 October 2026 and no longer showed medians — ChatGPT appears to have answered from an older copy of the page.
Finding 4: Google's AI and Copilot leaned on different pages
Google's separate Generative AI features report counted 702 impressions across 32 pages between 31 May and 24 October 2026. One page — the flood zones guide — took 455 of them. Copilot cited that same page twice. An impression and a citation are not the same measure, and the date ranges differ, so compare the order of pages, not the size of the numbers (Google's page counts also add up to a little more than the 702 total):
| Page | Google AI features impressions (31 May to 24 Oct) | Copilot citations (14 Jul to 24 Oct) |
|---|---|---|
| Flood zones guide | 455 | 2 |
| Gated communities | 37 | 14 |
| Talon Preserve community site | 32 | 16 |
| New construction vs resale | 25 | 6 |
| Palmer Ranch guide | 10 | 48 |
| Homestead exemption | 7 | 31 |
| Flat fee MLS | not listed | 31 |
The two AIs did not merely cite the same site at different rates — they built their answers out of different pages of it. A page can be a site's strongest asset in one engine and nearly absent in another, on the same day, from the same content.
What got cited, and what didn't
Putnam records the pattern in his own study, with his own caution attached: the pages Copilot didn't cite weren't compared against the cited ones, so these traits are a pattern worth testing, not proof of cause. Most of the cited pages share three traits:
- 1 question per page. Palmer Ranch, homestead, flat fee listings, a single community.
- The answer is in the page text. Sale figures, fees and steps are written out in the text, and most of these pages name where their figures came from.
- The figures are dated. The community sites give the date range their sales figures cover.
And the negative result: neither calculator page was among the cited pages Bing listed. In Google's AI features, the property tax calculator page was shown 4 times out of 702, and the payment calculator page wasn't listed. Putnam's hypothesis, which he states he has not tested: an AI answer can quote a sentence, but it can't run a calculator for the person asking.
He also notes the result partly matches a 2024 study from Princeton and IIT Delhi (GEO: Generative Engine Optimization, KDD 2024), which found its methods raised a page's visibility in AI answers by up to 40%, and that adding statistics, quotations and cited sources helped. That study tested its own set of questions and Perplexity; it did not test Copilot or ChatGPT.
What this means: AI citation is not Google rank
Putnam has now measured the decoupling from both directions. In his first study, 81 businesses that had already won the Google game — every one in the top ten of Google Maps for its trade and town — were mostly invisible to ChatGPT: 70% never named, and ChatGPT cited third-party directories instead of the businesses' own sites. In this second study, a site that has not won the Google game at all — average position 30.2, 72 clicks in 92 days — was cited 205 times by Copilot and placed first by ChatGPT, ahead of the largest real-estate portal in the country. High Google rank did not produce AI naming in the first study. Low Google rank did not prevent AI citation in the second. The two systems are measuring different things, and a dashboard that only shows one of them is blind to the other.
The second study adds a refinement the first could not: "AI visibility" is not one number. Copilot, Google's AI features and ChatGPT each leaned on different pages of the same small site — the flood zones guide carried Google's AI impressions and was nearly invisible to Copilot, while the Palmer Ranch guide did the reverse. A business measuring only whether it appears somewhere in AI answers, without asking which engine and which page, is averaging away the actual finding.
That per-engine, per-page measurement is the work ScriptMasterLabs does. The AI Visibility Report measures a business from the outside in — whether ChatGPT, Perplexity and Google AI name it, who they name instead, whether the rating they show matches reality, and which sources feed the answer — because Putnam's two studies together say the furnisher layer, not the Google ranking, is where local AI recommendations are decided. His first study is the evidence from the trades; this one is the same evidence from inside a single site's own console reports.
For readers of this study
This section is ScriptMasterLabs' catalog, not Putnam's — the study above stands on its own, and nothing here is part of its findings. Two of the study's audiences asked for tools like these:
For property investors reading a home-value study for the deal math underneath it: the Rental Property Deal Analyzer — a 6-tab Excel deal machine with 46 live formulas. Run any rental through Base, Conservative and Stretch scenarios and see monthly cash flow, cap rate, and cash-on-cash return instantly, with the downside case shown in red before the seller sees it. $19.99 at shop.scriptmasterlabs.com/products/rental-property-deal-analyzer.
For realtors and local-business owners who want Putnam's measurement run on their own business: the AI Visibility Report — your credit report, but for AI search. Does ChatGPT, Perplexity or Google AI name you for your service and town? Who got named instead? Is the rating they show the right one? Which sources feed the answer, and what is the fix list? $79 one-time report, $29/mo monitoring, at scriptmasterlabs.com/ai-visibility-report.
What the study does not claim
Putnam states his limits in his own words at the end of the study, and they are part of why this data is citable. They are reproduced here verbatim:
- "This covers 1 site. It's a case study, and a site in another market could see something different."
- "The 2 reports cover different dates. Bing covers 103 days (14 July to 24 October). Google covers 92 days (25 July to 24 October)."
- "Bing's report shows no data before 14 July 2026. Citations before that date, if any, aren't counted here."
- "The 2 AI counts measure different things. ChatGPT gives site owners no counts. Google's AI features report counts impressions, not citations, and covers a different date range. Bing's regular search report wasn't read. The real total across all AI answers is unknown."
- "Bing shows a sample of the questions. Of the questions behind the 205 citations, Bing showed 1: 'florida flat fee mls ai'."
- "Google hides most of the searches behind the 72 clicks."
- "The ChatGPT result is 1 question on 1 day, plus the August results from my notes. ChatGPT's answers change from one run to the next."
- "I build and maintain the site. I'm a licensed real estate agent with Putnam Realty Group, and the site brings me business. I've published the numbers as the reports showed them, with the dates, so anyone with the same reports for their own site can compare."
One limit ScriptMasterLabs adds on its own behalf: this is Michael Putnam's single-site case study in Sarasota County real estate, not ScriptMasterLabs' research, and the 205-citation figure should always travel with its source, sample, and dates attached — 1 site, 14 July to 24 October 2026, Bing Webmaster Tools AI Performance.
Frequently asked questions
Can a website Google ranks poorly still be cited by AI answers? Yes. In Michael Putnam's 2026 case study, his Sarasota County home-value site was cited 205 times by Microsoft Copilot between 14 July and 24 October 2026 while its Google average position over a similar stretch was 30.2. AI citation and Google rank moved independently.
What kind of pages did Copilot cite? Written-answer pages: guides and community pages with the answer written out in the page text. The most-cited page was the Palmer Ranch guide (48 citations), followed by the flat fee MLS page and the homestead exemption page (31 each). Neither of the site's two calculator pages was among the cited pages Bing listed.
Do Copilot and Google's AI features cite the same pages? No. The site's flood zones guide drew 455 Google AI-feature impressions but only 2 Copilot citations, while the Palmer Ranch guide drew 48 Copilot citations but only 10 Google AI-feature impressions. An impression and a citation are different measures over different date ranges, so compare the order of pages, not the size of the numbers.
Did ChatGPT recommend the site? Yes. On 26 October 2026, signed out, ChatGPT was asked which local website shows home values in Talon Preserve, Nokomis FL, and listed the site's Talon Preserve page first, above Realtor.com. In an earlier test on 19 August 2026, it listed the site first for three of four communities asked about.
How big is the study and who ran it? One website in Sarasota County, Florida, measured from five console and AI readings taken on 26 October 2026 by Michael Putnam, a licensed real estate agent with Putnam Realty Group in Nokomis, Florida, who builds and maintains the site himself and discloses that in his stated limits. Full study: https://floridahomevalueai.com/ai-visibility-study-2026
How can I measure whether AI cites my website? Read the same reports Putnam read: Bing Webmaster Tools' AI Performance report shows Copilot citations per page, and Google Search Console's Generative AI features report shows AI-feature impressions. ScriptMasterLabs' AI Visibility Report runs the outside-in version of that measurement — whether ChatGPT, Perplexity and Google AI name your business, who they name instead, and which sources feed the answer: https://scriptmasterlabs.com/ai-visibility-report
FOLLOW
X: @TimmyCrypto78 — https://x.com/TimmyCrypto78
Instagram: @scriptmasterlabs — https://www.instagram.com/scriptmasterlabs
Threads: @scriptmasterlabs — https://www.threads.net/@scriptmasterlabs
Telegram: @x402sml — https://t.me/x402sml
LinkedIn: https://www.linkedin.com/in/timothy-walton-23224052
MY LINKS
Website: https://scriptmasterlabs.com
Books: https://scriptmasterlabs.com/books
Shop: https://shop.scriptmasterlabs.com
Gumroad: https://timwalton.gumroad.com
Whop: https://whop.com/timmy741
Ko-fi: https://ko-fi.com/scriptmaster
Discord: https://discord.gg/jGWDutyVAR
Verdict Protocol: https://verdict.scriptmasterlabs.com
Free VA tools: https://va-ratings.org
Top comments (0)