Methodology

Where our stocking data comes from, how the text is made and checked, how often it updates, and why some pages are not indexed yet.

Where the data comes from

Stocking rows come from a state wildlife agency's own public report or published stocking plan, not a third party or a forum. For most states the rows are reports of fish already stocked. Arizona, California, and Oregon publish only forward plans, so every row there is an agency plan: it is labelled "Scheduled", is never shown as a last stocking date, and is not counted in stocking totals or recent-stocking lists. Iowa, Kentucky, Pennsylvania, and Texas mix completed stockings with plan rows; only the completed ones count. Right now the sources are:

Each stocking table on the site links back to the agency report it came from, so you can always check the source yourself.

How lake and species facts are added

Lake facts (surface acres, elevation, access) and species tactics come from agency species profiles, federal geodata, and other public research. Each one is stored with the link and date of the source it came from. Automated checks compare the fishing notes and lake overviews with their cited pages where the checker can read them, and what fails is removed rather than printed as a guess. That's why some lake pages show fewer facts than others: it means we don't have a checked answer yet, not that the fact doesn't exist.

How this content is made

  1. The data is sourced. Stocking records are pulled from state wildlife agency reports into a database, each with the source it came from and the date we pulled it. Lake facts (location, access, species found) come from named public sources and are stored with a source link and the date it was read.
  2. We check it. When a short fishing note is drafted, an automated check requires each claim to match a verbatim quote from a fetched page. A second check compares that page's title, coordinates, county and water type with the lake (a pond is not a lake, a creek is not a river) and rejects pages about a different water; it can still miss two waters that share a name. Each lake's map pin is also checked against its state and, where an agency publishes a county, against that county; a pin we cannot verify is not drawn, and the page shows no map, elevation or nearby lakes for it. On every build another check rejects notes that are about a different water, restate a fishing rule, carry a health, water-quality or safety claim (advisories, bacteria, algae, closures), present a one-off event, plan or ongoing operation as a standing fact, say a species is stocked when the site's own stocking record does not, or say nothing about fishing. For lake overviews, the checker compares numbers and years with the cited pages it can read. Some cited pages, such as PDFs and sites that block automated readers, cannot be read by the checker, so for those the link is there for you to check. We add spot checks on top, and text that fails a check is removed, not guessed.
  3. A language model writes the readable text. The short fishing notes on lake pages are drafted by an AI model from fetched, cited pages only, inside a strict schema where each claim is tied to a verbatim quote. Lake overviews, spots and tactics, species profiles, and guides are drafted by AI agents from the sources they cite, against a typed schema in which each field carries its own sources; those fields are not quote-tied, and audit scripts flag missing sources instead. Where you see such text, it is labelled: "Drafted by AI from the cited sources and checked before publishing."
  4. We keep auditing. The build runs automated content audits, and we audit published pages against their cited sources from time to time. When we find an error we fix the cause, not just the one page, and remove text we cannot support.

Some sentences are not drafted by a model at all: counts, dates, "last stocked" lines, and the lake, county, and state summaries are generated by software from the records, not written by hand. Illustrations without a photo credit are AI-generated and labelled as such.

The same process applies across the Outdoor Content Network: shared refresh cadence, indexability gates, and reviewer steps.

You can check the claims yourself: each AI-drafted passage sits next to links to the sources it was drafted from and the date we checked them, and every lake page lists all of its sources. If something is wrong, tell us and we will fix it.

How often this updates

An automated GitHub Actions job pulls the agency sources above every Monday and Thursday for 24 of 25 states (Utah is not in that job yet) and writes the new rows to our database. It does not build or publish the site: the site is rebuilt by its owner from the latest pull, so a page shows data as of its last pull, not live. Every stocking table links to the agency's own report for that event, so you can check the agency directly for anything more recent than what's shown here. A lake's most recent listed stocking is the latest completed plant in the data behind that build; a planned or scheduled date is shown as "Scheduled" and never as the last stocking.

Why some pages aren't indexed in search

A lake page indexes when it clears one of three bars (same rule as isLakeIndexable in the build): a destination guide of at least 3 combined named spots and species tactics whose claims are verified, or at least one confirmed stocking event (a planned or scheduled date is not one) plus either 4 different calendar years of confirmed stocking, or a verified access fact or species record (a dated agency source, or a claim checked against its page). Text that is not verified is not shown on the page and does not count. Elevation, acreage, and a photo do not clear that bar on their own. Pages below it still exist on the site and you can browse them directly, but we mark them so they don't show up in search until they've earned it.

This is deliberate. A site with thousands of thin, near-identical pages is bad for readers and gets penalized by search engines. We'd rather grow slower and stay useful.

Regulations

We aim not to restate fishing regulations, bag limits, gear rules, closures, or license rules. Those change and carry real consequences if we get them wrong, so check your state agency for current rules. The build runs an automated check that fails if a lake note, access note, destination guide or species note contains fishing-rule language such as limits, seasons, gear rules, or licences; guides are scanned for bag-limit numbers and species and guide wording is flagged for prohibitions. The check uses patterns and can miss a phrasing, so treat the agency's page as the authority. Physical access facts that a managing agency states on its own page, such as a ramp or a day-use fee, may stay, with the source link beside them. Fish-consumption, water-quality and safety claims (advisories, contamination, bacteria, algae, closures, drownings) are not carried in lake text at all: the build removes those sentences from every source tier, because they change and belong to the health or managing agency, and the page links the agency.

Full source log

We keep an internal log of every data source we use, its terms of reuse, and the date we last checked it, alongside the specific adapter or record it feeds. It's part of our build process rather than a page on the site, but the short version above covers what matters for readers.