Making a Weekly Racing Show With AI

A NASCAR discussion recorded on Tuesday has a short shelf life. By Saturday night, the predictions face a result. For Queen City Garage, the production job includes editing the show, checking names and statistics, finding supporting images and preparing coverage people can use before the race.

For the third episode of The Grand National Show, we used AI across several of those steps. It helped locate material in the transcript, draft companion articles, assemble graphics and prepare revisions. The practical question was how much of the work between the recording and editorial review it could handle—and how much checking its output would require.

One assignment began with Mamba Smith’s prediction that Carson Kvapil would win the NASCAR O’Reilly Auto Parts Series race at Gateway. The selection took about twenty seconds near the end of the recording. His explanation of the track came several minutes earlier: different demands at each end, patience applying the throttle and the straightaway speed lost to a poor corner exit.

To turn that into a column, those passages had to be found and connected. The other panelists’ selections had to be identified and their reasoning attributed. Race information and the points standings needed checking. A photograph had to be selected. The resulting argument needed to make sense to someone who had not watched the show.

Working from the transcript, AI located the relevant passages and proposed source ranges for excerpts. It helped draft the column, bringing the prediction forward and placing the track explanation behind it. It also assembled a comparison of the panel’s four selections and prepared supporting social copy.

That gave us a first version to edit. We could examine the selected passages and the proposed structure together, rather than begin each task with another search through the recording. The distinction matters: finding a passage is one job; deciding whether it supports the article is another.

The same approach extended to the graphics. The revised review package contained eighteen new boards, alongside the approved top-three standings reference. They covered driver spotlights, the Chase field, Gateway and the panel’s picks. AI helped assemble the layouts using actual photographs, logos and supplied race data, then produce full-frame previews for review.

Some of the most useful changes came after those previews were examined. A Jeremy Clements card had listed his race statistics without showing his photograph. Mamba wanted the driver on screen and more useful context. The revised card used a portrait, his ninth-place Darlington finish and his two career series wins.

On the top-three standings board, he wanted Carson Kvapil’s nineteen-point deficit emphasized alongside Sheldon Creed’s seven-point gap. He questioned the need for total points. We removed the totals and gave the two deficits consistent treatment.

Those corrections illustrate both the opportunity and the cost of the process. AI could help build the next version from specific instructions. It had not independently arrived at the strongest editorial choice. Someone still had to see the problem and articulate the change.

The transcript presented a different risk. Dillon Welch joined through Mamba’s phone, and the automated transcript placed the exchange under Mamba’s name. Taking that label at face value would attribute Dillon’s Brandon Jones pick to the wrong person. The phone contribution has to be checked against the recording before it becomes a published attribution or caption.

The written material required revisions as well. The first production articles spent too much time describing intentions. The racing column repeated parts of its track explanation. Review turned those into concrete instructions: build the production story around an actual correction, tighten the column and preserve the disagreement among the panelists.

The visible result of this work is a graphics review package, three companion article drafts, share images and source selections for clips. These are outputs at different stages, not a claim that the whole episode and every accompanying asset are finished. The clips still have to follow the approved edit, and the complete production needs playback review.

We have not run a timed comparison against producing the same package entirely by hand. There is no defensible percentage reduction or hours-saved figure to attach to this episode. The evidence is narrower: AI handled portions of the searching, drafting, layout preparation and revision work, while the production team supplied direction and evaluated the results.

For a weekly show, that is a useful place to examine the economics. A faster first draft creates value only if the checking and correction leave enough time to finish something worth publishing. The number of files generated is less important than how many become usable before the race.

At QCG, that is the next test: carry the material through final review, record the effort required and see which parts of the process should become routine for the following episode. AI has helped put more of the work in front of an editor. Getting it through that editor and to the audience remains the production job.

Follow The Grand National Show at Grand National Today. Read about our first contributor packages in The show is not done when it publishes.