Skip to content
specpair

Editorial Policy

Last updated: November 21, 2026

SpecPair is made by Zeplik, Inc. Its spec data, verdicts and articles are AI-assisted. This page explains where the specifications come from, where AI is used, what is checked before a page is published, and how to report a mistake.

What this covers

This policy covers the product, comparison, category, brand, roundup, news and blog pages on specpair.com, the product advisor, and the articles from our content pipeline that appear in the blog, comparison, best-of, alternatives, glossary and data sections.

Where spec data comes from

  • Research. When a new product is announced, an automated research step fetches the page that announced it, the maker’s own specification page where one can be reached, and the product’s entry in established spec databases and publications, such as GSMArena for phones, Notebookcheck for laptops and SoundGuys for headphones.
  • Extraction. An AI model (Claude by Anthropic) reads those pages and fills in only the values it finds in them. It never guesses a missing value, and it records which source each value came from. A draft with too little data and no price is rejected automatically.
  • Approval. A person approves each new product in SpecPair’s admin before it appears on the site.

How scores are calculated and how a winner is picked is explained in our methodology. Comparison pages show a “Last verified” date for their spec data.

How we use AI

  • Verdicts. Written verdicts on comparison pages are drafted by an AI model (Claude by Anthropic) from the specifications stored for each product, and must cite those figures. Two-product pages without a written verdict carry a summary generated from the spec sheets, and say so on the page.
  • The advisor. The product advisor answers with an AI model (Claude by Anthropic) working from SpecPair’s own product catalog, roundups and blog posts.
  • The site itself. SpecPair’s code, page templates and tests are written with an AI coding assistant.

We do not invent statistics, quotes, studies, testimonials, user counts or results.

How articles are made

Articles from our content pipeline (guides, comparisons, best-of lists, alternatives, glossary entries and data studies) go through these steps:

  • Brief and draft. An AI model writes a brief for each article from search data and the pages that already rank, then drafts the article from that brief. Every brief must name what the article adds that the existing results do not, such as our own data, test results or a worked example.
  • Automated checks. A script checks each draft: every claim must have a source, banned filler phrases fail the draft, and the page structure, links and images are checked.
  • Fact check. A separate AI run with web access checks every claim against a live source. It sees the draft and its sources, never the writer’s reasoning, and the run that wrote a draft cannot fact-check it. Titles, descriptions, structured data and image alt text are checked the same way. A claim that fails is fixed or removed, and the check runs again.
  • Approval. Each article is approved before it is published: by the site’s owner, or, when it passed the automated checks and the fact check with no open flag, automatically under rules the owner set. Pages on health, money, legal or safety topics, and changes to the title, heading or address of a page that already ranks in search, always wait for the owner. The owner reviews a random sample of each week’s pages, every automatically approved article is part of that review, and any automatic approval can be undone.

AI-generated images may illustrate an article. They never show real products, people, results or evidence.

How we test products

SpecPair’s own comparison pages and roundups compare published specifications. They are not hands-on reviews, and they do not claim to be. An article from our content pipeline that compares, ranks or lists alternatives to products is published only when:

  • we ran every product on the page ourselves, with the same inputs;
  • the method, date, inputs, outputs and dated screenshots of each product are on the page;
  • every “best for” pick says why, with that evidence;
  • the page says near the top that Zeplik makes any of its products that appear, and says plainly where a competitor wins.

SpecPair earns money from display advertising (Google AdSense). No one pays to change a score, a winner or a ranking, and rankings are never ordered by commission or by which product is ours.

Sources and dates

Every fact in an article links to its source, primary sources first: official documentation, standards bodies, maker pages and government sites. Prices and specifications carry an “as of” date. A page’s “Updated” date and its sitemap date change only when its content, structured data or links change in a way a reader would notice.

Corrections

If you find a wrong specification, score or claim, email [email protected] or use the contact page, with the page and what is wrong. We check it against the source. A wrong specification is corrected in the product’s record, which every page that uses it reads, so the product page and all of its comparisons change together; a written verdict that rested on the old figure is drafted again from the corrected data. A wrong claim in an article is fixed or removed, and the corrected article goes through the same checks as a new one before it is published again.

Contact

Zeplik, Inc.
1111B S Governors Ave
Dover, DE 19904
[email protected]