What JAC's own search checks revealed
A dated baseline shows AI impressions, an outdated service description, and the next checks worth running.
The short version
- JAC recorded 24 Google generative AI impressions from September 9 through October 6, 2026. This is a baseline, not a growth result.
- One branded ChatGPT answer used retired service labels. One unbranded provider answer didn't name JAC. These are two distinct spot checks.
- A Google live test can pass while an index record says the URL is unknown. Both checked pages were indexed that evening.
- Use a fixed question and documented settings for a repeat check, then inspect accuracy before celebrating a mention.
JAC’s first search checks revealed an outdated AI description, a small Google AI impression baseline, and two important pages initially missing from Google’s index. Both pages were indexed by the evening follow-up. Those findings point to work we can inspect. They don’t yet establish that the recent site refresh improved visibility or generated a lead.
This log records the starting evidence for JAC’s own website. It gives B2B SaaS founders a concrete example of how to assess their site before interpreting an AI visibility dashboard. The questions, settings, dates, and limits matter as much as the counts.
What did the Google report show?
On October 9, 2026, JAC’s native Search Console Generative AI features report showed 24 property impressions for September 9 through October 6. Web was the selected search type. The report said its last update was 32 hours earlier.
That is a Google generative AI impression count for the selected window. Google’s report documentation describes coverage of AI Overviews and AI Mode. The count doesn’t tell us which individual buyer asked a question, whether they visited, or whether a sales conversation followed.
Nine page rows appeared in that window. They included the homepage, Services, and several educational pages. The longer July 7 through October 6 window showed 50 property impressions and 14 page rows. Its page-row counts summed to 51, so the property chart and page table can’t be substituted for each other. The report documents their different aggregation methods.
| Native report record | Value |
|---|---|
| Report | Generative AI features, Web selected |
| Completed date window | September 9–October 6, 2026 |
| Property impressions | 24 |
| Page rows returned | 9 |
| Observed on | October 9, 2026 |
The baseline ends before the latest content batch. No matched later window has been assessed. Calling these impressions a refresh result would reverse the order of the evidence.
What did the branded ChatGPT check reveal?
The first question was:
Search the web. What services does JAC Growth Marketing offer to B2B SaaS founders? Cite the sources you use.
This was a fresh temporary consumer chat with Unpersonalized selected. The visible setting said it would ignore memory, plugins, and custom instructions. The interface displayed Medium; the underlying model version and location weren’t displayed. The prompt was in English and explicitly requested web search.
The completed answer named JAC, displayed links to owned pages, and used retired service labels. Its published prices matched the current offers. The verified issue was the service description, including superseded positioning.
That matters because a buyer researching a company needs to understand what it currently offers. A source link beside an answer doesn’t establish that the answer reflects the page’s current text. We saved the complete answer and displayed links before checking material statements against the current service pages.
| Branded answer check | Reviewed finding |
|---|---|
| Names JAC | Yes; the question supplied the name |
| Displays owned links | Yes; including home and Services |
| Service labels | Retired labels; description mismatch |
| Published prices | Match current offers |
| Recommendation | Not assessed |
The current names are AI Search & SEO, AI Audit, and Founder-Led Content. Search is the main monthly offer, the audit is an optional one-time assessment, and founder content includes writing, scheduling, posting, and reporting. The audit delivery guide explains the standalone scope.
The mechanism behind the stale answer remains unknown. We haven’t established whether cached material, another source, or answer generation produced the old description. OpenAI’s crawler guidance distinguishes search crawling, training, and user-triggered visits. Those differences help frame an investigation; they don’t diagnose this answer by themselves.
Did JAC appear in an unbranded provider answer?
The second question was:
Search the web. Who can help a small B2B SaaS company improve visibility in AI search and handle its search pages? Cite sources and explain how to compare the providers.
This was another fresh temporary, unpersonalized consumer chat with the same visible model display and an English search request. The completed answer compared other providers and didn’t name JAC.
That is one observed absence for that question and context. It isn’t a claim that JAC never appears, a market-share estimate, or a measure of buyer demand. The question was a research hypothesis based on the current offer, rather than a verified customer query.
The two checks also have different jobs. Supplying JAC’s name tests its description. Leaving the name out tests an instance of discovery. The ChatGPT diagnosis article explains why a branded answer can’t substitute for the second task.
Could Google fetch the current search pages?
The Google index records for the monthly search-service page and the full search guide said their URLs were unknown. Neither showed a last crawl. The property’s Search generative AI setting showed Include, so we left it unchanged. Live tests returned a different, narrower finding:
| URL | Live test on October 9, 2026 | Index record at inspection |
|---|---|---|
/ai-search-seo/ | Available to Google; can be indexed, 7:20 p.m. Eastern | Unknown; no last crawl |
/guides/aeo-for-b2b-startups/ | Available to Google; can be indexed, 7:22 p.m. Eastern | Unknown; no last crawl |
Indexing requests for both were accepted into the priority crawl queue. That confirms submission, not indexing. A later inspection needs to establish whether either status changed.
Google’s current AI optimization guide ties eligibility to indexing, snippet eligibility, and inclusion in its Search generative AI features. It doesn’t require special AI markup or a new page for every variation of a question. Publishing another file can’t replace checking those conditions.
What changed in the evening follow-up?
A later read-only Google index inspection reported Submitted and indexed for both the search service and full search guide. Each returned an October 9 crawl date. The newly published Tampa comparison was still unknown to Google at that inspection.
| URL | Earlier October 9 index record | Evening index record |
|---|---|---|
/ai-search-seo/ | Unknown; no last crawl | Submitted and indexed; October 9 crawl |
/guides/aeo-for-b2b-startups/ | Unknown; no last crawl | Submitted and indexed; October 9 crawl |
/blog/tampa-bay-marketing-agencies/ | Not in the earlier inspection | Unknown; no last crawl |
The evening result establishes an index record for two pages. It doesn’t establish that either page appeared in an AI answer, attracted a visit, or generated an inquiry. It also doesn’t establish why Google indexed them.
Both exact consumer questions were repeated in separate fresh temporary, unpersonalized chats in English. The visible setting remained Medium. The branded answer still used retired service labels, and the unbranded answer again didn’t name JAC. Displayed owned links in the branded answer included older routes that redirect to current pages.
The initial answer records stay intact. These same-day repeats are additional spot checks, with the underlying model version and location still unknown. They aren’t a completed fixed panel or a visibility growth rate. The Google reporting window hasn’t been compared with a matched post-change period.
The next repair question is narrower now. Check whether the sources used in a stale answer reflect current page text. A displayed link alone doesn’t resolve that question. Keep the redirects, compare current descriptions, and record what the next completed answer says.
What happens after this baseline?
The controlled work is to keep current offer facts clear on the pages the answer cites, make useful search guidance easy to find, and record the indexing state. Record each publication date so a later observation can be placed in context.
The next comparison will repeat the exact questions again after a later observation window, save the answers and displayed citations, and review material accuracy. Changes in the interface, hidden context, or answer variation will remain part of the record. These initial spot checks aren’t the completed repeated panel needed for a comparable rate.
Native report periods, consumer observations, referral visits, and qualified inquiries will stay separate. The measurement guide explains the denominators and missing-data rules. When a new comparable window exists, this log can report what changed and what still can’t be attributed to the site work.
Start your own record with one buying question and one saved answer. The free AI visibility baseline kit supplies the sheets and this completed example. The AI search overview connects the diagnosis to page work. If you’d like research, implementation, and reporting handled together, explore AI Search & SEO and book a call to discuss your priorities.