Reading the monthly report
What a GEO Report Should Show, and the Red Flags in Bad Ones
The report lands on the fifth of the month. Your AI Visibility Score rose from 61 to 68. Nobody can say what the score measures, which questions produced it, or what anyone did to move it. You paid for that number, and you deserve better from the next one.
October 18, 2026 · 9 min read
The short answer
A monthly GEO report should show the actual AI answers to a fixed set of customer questions, whether you were named and cited, who was named instead, which sources the engines used, what was fixed that month, and first-party data from Search Console, Bing Webmaster Tools and GA4. A single score with no answers behind it is not a report.
On this page
A score is a summary, not a report
Scores are not the villain. A single figure is a handy way to watch a trend. The problem is a score with nothing underneath it. AI visibility is made of specific answers to specific questions, and an owner should be able to read some of them: what ChatGPT said when asked for a roofer in Seguin, which businesses it named, which pages it cited. If a report cannot show you that, it is reporting a mood.
A good report also ties measurement to work. GEO is not a weather forecast you subscribe to. Each month something should have been fixed (a listing, a page, a schema error, a backlog of unanswered reviews), and the report should say what, and whether the answers moved afterward. Google says indexing and serving are never guaranteed, and trade press reported in February 2026 that Bing's updated Webmaster Guidelines say the same of GEO and citations. A report that implies a guarantee is selling something other than GEO.
A visibility score without the answers behind it is a horoscope with a logo.
A sample outline, section by section
This is the outline we would expect from any competent monthly GEO report, ours included. Some sections run a page; some are a single line. The order matters: answers before scores, work before plans.
| What it shows | Where the data comes from | What to look for | |
|---|---|---|---|
| 1. Summary | Three to five sentences on what changed, what was fixed and what comes next | The people who did the work | Plain language and specific claims; no unexplained score |
| 2. The prompt set | The exact questions tracked, the engines, the locations and how often each ran | The tracking method | The same questions as last month, with any changes noted |
| 3. Answers that matter | Full text of key answers, who was named, what was cited | Manual runs or a tracking tool | Your top buying questions, not your brand name |
| 4. Mention rate and share of voice | How often you and each competitor were named, by engine and question type | Counts from the answer log | Sample size stated; movement judged over a quarter |
| 5. Cited sources | The pages, directories and review sites engines quoted | Citations inside the answers | Sources you can fix, claim or earn |
| 6. Accuracy check | Wrong hours, prices, services or claims, and the status of each correction | Brand fact prompts | Every error has an owner and a date |
| 7. Google first-party data | AI impressions by page; Web clicks and impressions | Search Console | Pages gaining or losing AI impressions |
| 8. Bing first-party data | Copilot citations, grounding queries, Citation Share | Bing Webmaster Tools | Grounding queries you did not expect |
| 9. Visits and leads | AI assistant sessions, landing pages, key events; Business Profile calls and direction requests | GA4 with a custom AI channel and a linked Business Profile | Leads, not just sessions |
| 10. Work completed | Listings fixed, pages published, schema corrected, reviews answered | The work log | Items you could verify yourself in ten minutes |
| 11. Next month | The planned fixes and the reason for each | The team | A short list tied to what the answers showed |
Sections 2 through 6 come from asking questions and reading answers; our guides to prompt tracking and AI share of voice cover the method. Sections 7 through 9 come from accounts you already own. Section 10 is the one most often missing.
Google's and Bing's own numbers belong on page one
Third-party tracking estimates visibility from outside. Google and Microsoft now publish data from inside their own engines, and a report that skips it is ignoring the only figures that come straight from the source. All four of the instruments below are free.
Search Console, Generative AI performance
Every site has had this report since August 31, 2026. It counts impressions only: how often links to your site appeared in AI Overviews and AI Mode, by page, country, device and date. Search Labs experiments are excluded and the usual 1,000-row limit applies.
Search Console, Web performance
Clicks from AI Overviews and AI Mode are counted here under the Web search type, mixed in with ordinary results. An AI Overview occupies one position, and every link inside it shares that position.
Bing Webmaster Tools, AI Performance
In public preview since February 10, 2026: citations in Copilot, Bing's AI summaries and some partner integrations, plus grounding queries and page-level citations. Intents, Topics, Citation Share and Compare arrived on June 16, 2026. No clicks.
GA4 and Business Profile
The AI Assistant channel (live since May 13, 2026) plus a custom channel for the visits it misses. Since June 8, 2026 GA4 can also show linked Business Profile calls, bookings and direction requests.
Bing's grounding queries deserve special attention. They are the phrases Copilot used to retrieve your pages, the closest thing any engine offers to the questions behind its answers. A good report reads them each month and feeds new ones into the prompt set. The setup for the analytics row is in how to track AI traffic in GA4.
Red flags in an agency GEO report
- A proprietary score with no method. If nobody can say which questions, which engines and how many runs produced the number, it cannot be checked or compared with anything.
- No prompt list. You should be able to read every question tracked. Hidden prompts tend to be easy prompts.
- Mostly branded questions. Engines describe almost any business when asked about it by name. Visibility on your own name is nearly free.
- "Ranked number one in ChatGPT." AI answers are generated fresh and vary between runs, and there is no fixed number one. A screenshot of one good run proves one good run.
- Claims of special access. Google says no third-party tool has access to its internal ranking or AI systems. Anyone implying a back door is overselling.
- llms.txt or FAQ markup reported as wins. Google says Search does not use llms.txt, and it stopped showing FAQ rich results on May 7, 2026. Neither is a result worth a slide.
- AI Overview traffic counted twice, or not at all. Those clicks sit inside Organic Search in GA4 and inside Web in Search Console. The report should say how it handled them.
- No work log. Months of charts with no list of what was fixed means you are paying for observation.
Real progress versus a flattering chart
Signs of real progress
- The same questions, asked the same way, name you more often across a quarter
- Wrong facts found in earlier months are now answered correctly
- New sources cite you: directories, local press, industry sites
- Search Console shows more of your pages earning AI impressions
- Bing's grounding queries widen to cover more of your services
- Calls and forms from AI and organic visits rise together
Signs of a flattering chart
- The score rose, and so did the number of easy prompts
- Gains appear only on questions that include your name
- One month's jump is celebrated with no sample size given
- AI Overview clicks are folded into AI totals with no explanation
- Nothing in the work log explains the change
Movement takes time, and the clock differs by engine. OpenAI says robots.txt changes reach its search systems in about 24 hours, while a corrected listing or a new review profile can take weeks to appear in answers, and some engines barely shift for months. Judge trends by the quarter. When a report shows months of nothing, the fix pages are a good place to check whether the program is aimed at the right problem.
What a DataCram monthly report contains
Our reports start with the answers, not a score. Each month a client sees the questions we track, the answers the engines gave, who was named instead, the sources cited, the errors found and corrected, the Search Console, Bing and GA4 figures, and the list of work done. Monitoring runs on AICiteBeam, the tracker we built. On sites we build, the Athena CMS agent Delphi adds a weekly report: it asks ChatGPT, Google's AI answers and the others what they say about the business, flags wrong hours, prices or claims, and drafts the facts and FAQs engines cite, for the client's approval.
GEO is part of DataCram's monthly SEO programs. They start at $625 a month, most local businesses pay between $625 and $2,250, there is no long contract, and SEO financing is available. The tier is set after a free GEO audit. GEO pricing explains what changes the number, AI visibility monitoring describes the tracking in more depth, and our generative engine optimization guide gives the wider picture.
Questions owners ask us about AI search
Monthly is standard and usually right. AI answers vary from run to run and engines update on their own schedules, so weekly reports mostly show noise. Ask for a quarterly review as well, where the prompt set itself is checked and trends are judged across three months.
Mention rate and share of voice on a stated prompt set, cited sources, accuracy errors and their status, Search Console's Generative AI impressions, Bing's AI Performance citations, GA4 visits and leads from AI referrals, and a log of work done. The answers themselves should be readable, at least for the key questions.
Partly because AI answers vary between runs, so small samples swing several points on their own. It may also change because the prompt list, the engines or the method changed. Ask for the sample size and whether the questions matched last month's. AI share of voice walks through the math.
Partly. Clicks from AI Overviews and AI Mode are counted in the Performance report under Web, mixed with ordinary results. The Generative AI performance report, available to all sites since August 31, 2026, separates out impressions in those features, but not clicks.
Both, read together. Mentions show whether engines name you; traffic and key events show whether that brings business. Many AI answers produce no click, so traffic alone undercounts, and mentions alone cannot prove revenue. For local firms, phone calls and Business Profile actions often tell the clearest story.
Check the method first: same questions, same engines, enough answers. Then read the cited sources and accuracy errors, which usually explain why competitors are named. If months pass with no fixes in the work log, the program is watching rather than working. A GEO audit or a session of GEO consulting gives a second opinion.
Keep reading
- AI visibility trackersGEO Tools Compared: What Each Tracks and What It CostsSemrush, Ahrefs Brand Radar, Profound, Otterly.AI, Scrunch, Rankscale and more: what each GEO tool tracks and its published price as of October 2026.
- The AI engine mapAI Search Engines: The Models Behind Each One, and the Indexes They Actually ReadA map of the AI search engines: ChatGPT, Google AI Mode, Gemini, Perplexity, Copilot, Claude and Siri, the LLMs behind them and where they get sources.
- Search, for clearing crewsLand Clearing SEO: Get Found by the People Who Own the AcresSEO for land clearing companies: what landowners search, Business Profile setup, county pages that are not doorways, content that wins, and the first 90 days.
- Engine guide: ChatGPTHow to Get Recommended by ChatGPT: The Crawler, the Bing Connection and the Short ListHow ChatGPT search picks the businesses it names: the OAI-SearchBot rule, the Bing link, the sources it cites, shopping, and how to track the visits.
- Engine guide: AI OverviewsHow to Get Cited in Google AI Overviews, the Summary That Sits Above Your RankingWhat decides which pages Google AI Overviews cite, what Google says does not matter, how to measure it in Search Console, and how to opt out if you must.
- Concept: GEO and SEOGEO vs SEO: What Is Actually Different, and Where Google Says Nothing IsGoogle says optimizing for its AI answers is still SEO, and on Google it is. Here is what GEO adds for ChatGPT, Claude, Copilot and Siri, and what's folklore.
- Everything in Generative Engine OptimizationSee the section
Get a second opinion on your GEO report
Send us your latest report and the questions that matter to your business. We run them through the AI engines, compare what we find with what you were told, and explain any difference.
Which of these sounds like you? (pick any)
Rather talk now? Call (210) 346-0848.