An AI share of voice report template: measured in Linkeddit, designed in Canva

The core of the report is one number per engine: how often your brand is named, compared with every other brand named in the same answers. The templates that rank for this query agree on the shape. ai-visibility-software.com gives a nine column spreadsheet (prompt, engine, run date, brand mentioned, position, sentiment, competitor mentions, cited URL, notes) and the formula mentions divided by all mentions in the competitor set, times 100. Its worked example turns 28 mentions against 41 for a rival into 31.1%.

Riseklix adds the rest of a reporting page: citation coverage, citation share, answer accuracy and Search Console impressions, and notes that a brand can be mentioned often and still be described poorly. So a useful report has share of voice, presence, position and cited sources side by side, not one blended score.

Linkeddit's definition is strict about the denominator. A vendor named anywhere in a sample counts once for that sample, only products count, and the shares add up to 1 across vendors, so a long product list cannot inflate a score. That is the same idea AuthorityTech calls an open denominator, where brands you did not list still count against you.

A template tells you what to record. It does not collect anything. Someone has to ask each engine each question, read each answer, and type the names into the columns. AuthorityTech's protocol says one response is one observation, not a stable measurement, and ai-visibility-software.com recommends averaging at least five runs per prompt. By hand that is hundreds of rows before the first chart.

Linkeddit collects the runs on a schedule and returns the counts. On the test project, ChatGPT named the brand in 16 of 18 admissible samples, and the brand held 16 of 78 vendor votes, a 20.5% share of voice. Gemini gave 17 of 113, or 15.0%. The same brand scores differently on different engines, which is why the report is per engine.

Linkeddit supplies the data and does not write the analysis. Claude reads the figures and decides what to highlight, and you decide what goes to the client or the board.

Canva's announcement of the Claude connector says users can apply their Brand Kit when generating designs with Claude, so the first draft already uses their colours and fonts. The connector listing shows 17 tools, 12 reads and 5 writes, including generate-design, export-design, search-designs and add-comment-thread-to-design. Claude can build the report, you can comment on it, and Claude can export it.

Keep the division of labour clear. The figures come from Linkeddit, the layout comes from Canva, and the prompt tells Claude not to change a number. One viewer on a YouTube video about automating Canva posts with Claude wrote that great content cannot simply be automated. For a report that goes to a client, read every slide against the source numbers before you export.

Linkeddit measures through official provider APIs with grounded web search, and every result says so: it is a controlled measurement proxy, not an exact copy of what a person sees in consumer ChatGPT or Google. Put one line to that effect in the footer of the report.

Show gaps in the data as gaps. In the live run Perplexity returned errors for all 21 samples, so it has no figure, and Google AI Overviews appeared in 16 of 21 samples, with 5 showing no overview at all. A zero there would be wrong. A dash with the reason is right.

Finally, compare like with like. Change the prompt set, the engines or the competitor list and the trend line is no longer a trend, a point AuthorityTech makes about its own counting rules. Record the frame on the first slide.