Supermetrics MCP in Claude: pair it with Linkeddit for AI visibility and rank tracking

Supermetrics' Claude documentation says you can ask Claude to pull data such as spend, clicks and conversions from connected accounts, build dashboards and share them as artifacts or in Supermetrics Studio. It also lists campaign management for Google Ads, Facebook Ads, Microsoft Advertising and TikTok Ads.

The connector listing in the Agent Index shows 14 tools with read and write permissions. The data tools include data_source_discovery, accounts_discovery, field_discovery and data_query. Because it can write, keep these workflows to reads: nothing on these pages changes a campaign or a dashboard.

Linkeddit measures AI answers itself. Answer Radar asks ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews your buying questions on a schedule and records who is named, who is recommended and which domains are cited, with the sample counts behind every rate. Search records weekly Google ranks with the AI Overview shown above them, and Compete tracks what competitors change. All of it comes with one plan at $60 a month.

Linkeddit supplies data and writes no advice, so Claude does the analysis and the two connectors stay separate: Supermetrics holds your connected accounts, Linkeddit holds its own measurements.

Supermetrics does not compute share of voice inside AI answers, and its AI visibility guide says so. It describes prompt-level presence as something you measure by running a fixed set of prompts, sampled three to five times per engine, and says a third-party tracker's numbers are directional because sampling methods vary.

Linkeddit's answer_results_get returns share of voice per engine, where a vendor named anywhere in a sample counts once however many products it lists. On the test project in 2026-W44, the brand held 0.205 of ChatGPT's vendor visibility, 16 of 78 votes. Ask Claude to put that beside Supermetrics' Search Console impressions and GA4 AI Assistant sessions. For the weekly report, ask Claude to put answer_results_get with view=trends and interval=week beside Search Console impressions by page and GA4 sessions in the AI Assistant channel, one table per week, and to say the GA4 figure is a floor.

The need is stated openly. An SEO consultant on Reddit asked: "I’m curious how other SEOs are tracking AI visibility without turning it into a spreadsheet of manually checked prompts." Linkeddit records the questions and answers on a schedule, and Supermetrics adds the traffic side, so neither half is a manual spreadsheet.

Two, and both are covered on this page. The AI visibility workflow joins Linkeddit's measured answers, page watches and share of voice with Supermetrics reach data. The rank tracking workflow joins Linkeddit's weekly Google ranks and AI Overview flags with Search Console data pulled through Supermetrics.

The Linkeddit figures below come from read-only runs on Linkeddit's own test project, plausible.io, on 2026-10-29, so you can see the real fields Claude works from. The Supermetrics side is described from its documentation and connector tool list, not run on a live Supermetrics account.

Supermetrics' guide says the same prompt run once can return two different answers, and advises sampling each prompt three to five times per engine in a logged-out session and recording the rate, not the outcome. A screenshot of one run, it says, is an anecdote.

Linkeddit shows its sample counts. In 2026-W44 on the test project, ChatGPT named the brand in 16 of 18 admissible samples and Gemini in 16 of 20. That week's sampling was 3 runs with 1 sample per prompt, which is fewer per prompt than Supermetrics recommends, so read the rates as direction.

Appearance matters as much as presence. Google showed an AI Overview in 16 of 21 samples, an appearance rate of 0.7619, and the five absent samples are excluded. A presence rate quoted without that figure overstates how often buyers see an Overview.

Citation share lists the domains the engines cited. In Google AI Mode in 2026-W44, the brand's own domain and a competitor's each held 0.121, and reddit.com held 0.061. That is reddit.com as a site engines cite, which is worth knowing when you plan outreach. Supermetrics' guide notes that assistants name brands more often than they link them, so a citation-only view can say you are invisible where you are recommended.

Teams already do this by hand. An ecommerce SEO on Reddit described the routine: "Right now I’m checking a fixed list of buying questions every month, noting whether an AI Overview appears and who it cites, and pairing that with branded search and GA4 referrals from AI platforms." The same method runs here on a schedule, with traffic set beside it.

When an engine returns nothing, say it could not be measured. In 2026-W44 Perplexity returned no admissible samples on the test project: 21 of 21 requests errored. Linkeddit lists it as unmeasured with the reason, because a zero would read as invisibility. State the Supermetrics limits too: GA4 files assistant visits that send no referrer under Direct, so the AI referral count is a floor, and Search Console's AI report carries impressions with no clicks, click-through rate or query data.

Use answer_verify_get with kind=page. It lists each page watch with the day it changed and the day it will be measured. Two pages on the test project changed on 2026-10-27 and are due on 2026-11-19, 21 days later. Until then the verdict is null, which means not measured yet, not no change.

The result is an observed difference between equal windows and does not say the change caused it. Save the weekly prompt and rerun it after each measurement, so you compare a series and not a single reading.

Search Console reports average position across the searches where your page appeared. A rank check looks at one results page at one moment and records where you stood. Supermetrics' SEO guide lists tracked rankings as a job for tools such as Semrush or Ahrefs and warns that their data can be sampled or estimated, while Search Console is first-party.

So the two disagree on purpose. A keyword can average position 7 in Search Console and show as not in the top 10 in a single check, because the average includes searches where you ranked higher. Putting both on one row shows which reading you are looking at.

When two sources disagree, the argument is hard to settle without your own history. An SEO at an agency on Reddit, describing a page that kept dropping out of the rankings, wrote: "Using Accuranker and they confirmed it's not an issue their side." Dated rows from one consistent check give you something to compare other tools against.

Ask Claude to call search_visibility_get with view=ranks for every tracked keyword, query Search Console through Supermetrics for clicks, impressions, CTR and average position by query and page over the last 28 days, and build one row per keyword. Then flag falling clicks with a steady rank, an AI Overview that does not cite you, and queries at positions 5 to 15, which Supermetrics' guide describes as ones a small push could move. Ask it to label which connector gave each number.

On the test project, google analytics alternative ranked 2 and its AI Overview cited the site. For cookieless analytics the site was not in the top 10 (a rank of null, which is a result and not a missing check) and the Overview still cited it, so rank alone would have called it a miss. A citation is an observation about one check, not a promise of clicks, and Supermetrics' guide says Search Console's generative AI report carries impressions but no clicks.

Rerun it weekly, and use search_visibility_get with view=history, up to 52 weeks, for any keyword that moved. Search Console keeps 16 months of data, so older history needs exporting. The aio_impact view needs Search Console connected to Linkeddit, and on the test project it is not, so it returned a message asking for that and no numbers.