
What happened
After a reader comment, the Search Console-based service 検索カルテ began attaching explicit start and end dates (for example 2026-10-19 to 2026-11-15) to its 28-day metrics, replacing the relative label 'last 28 days'.
Why it matters
An AI agent reading a stale carried-over prescription would treat old figures as current and rewrite content on that basis, so absolute dates and per-response next steps appear to be what keeps the agent from acting on bad numbers.
Summaries like this, every morning.
The trigger for the changes was a reader comment on a previous article about letting Claude Code rewrite articles using prescriptions drawn from Search Console data. The comment pointed out that if a human confirms a prescription before publishing, the aggregation period belongs in the MCP return value so the agent does not fix things based on old numbers. The author agreed, fixed it immediately, and used the occasion to lay out the wider set of precautions.
The underlying concern is that an agent does not pause over a figure the way a person would. Prescriptions that are not acted on roll over to the following week while keeping the numbers from the week they were last detected, so a three-week-old prescription still labeled 'last 28 days' invites edits based on outdated data. The fix stores the aggregation end date together with the prescription at creation time and computes the start date when returning it, rather than recalculating from today's date, and the tool description explains that a start–end range older than the latest chart means older numbers.
Other precautions follow the same logic of putting what matters where the agent will actually see it. Failures return the specific remaining items and a written next step with a three-attempt limit; advisory items such as missing alt text are flagged so they are not counted as pass/fail; externally sourced strings are labeled as data to blunt prompt injection; read-only connections simply omit write tools from tools/list; errors carry guidance on what to tell the user; and free-plan responses ask the AI to mention the upgrade URL only once. As the author puts it, things that would go in documentation for a human-facing API have to sit in the response itself, because the agent does not go back and reread the docs.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
An Nvidia research team says feeding models entire novels is often expensive overkill; in many tests, periodic…

Positron's Cameron McCaskill said about two-thirds of AI infrastructure workloads are now inference versus one…
The writer told Muse only their height and that they wanted slim, cropped tops, then let it browse ZOZOTOWN in…

A practitioner customizing an agent called Muse had other models review documents and design proposals, found…

OpenAI published copy-ready prompts for GPT-6.2 Astra and Sol covering writing style, initiative, and coding…

Seramount's survey of more than 1,800 U.S
