AI article writer: Research to checked draft, in one run.

Two to four minutes end to end, and you can watch every step of it. The research pass, the four drafting passes, the images, the validation suite and the citation checks each announce themselves as they run. You can close the window; the run keeps going and the feed picks up where you left it.

A question AI assistants search when they answer about your category, taken from the AI visibility scan that onboarding runs, or failing that a keyword your competitors rank for and you don't. It's drafted in your voice once setup finishes and lands as a draft like any other.

No. A generation run ends in a draft in your Articles tab, and on a connected repository it also opens a pull request. Both wait for a person. The generator itself has no path to a live page: every route that changes your site is an action someone takes in the editor, after reading the draft. For what that review involves in one regulated field, see using AI to write health insurance content. The same question is answered for financial advisors, attorneys, CPAs, medical practices, property managers, insurance agencies, life insurance agents, Canadian life insurance agents and real estate syndicators.

The fact-verification standard operating procedure in the prompt tells it to omit the claim or soften it to general framing rather than publish an unverified specific. Separately, every external citation in the finished draft is fetched over HTTP and checked against the banned-source list for your compliance packs. That result becomes the Accuracy meter, so you can see how many of the sources actually resolved. To see how one of your existing pages reads to the assistants, run it through the free AI content readiness checker.

Guides, how-tos, comparisons, lists, and “alternatives to [competitor]” lists. The format is suggested from what the top 10 search results are, and you can change it in the brief. The writing follows the search intent too: a search to learn gets the answer first, a search to compare gets the criteria and honest trade-offs up front, and a search ready to act gets the specifics and one clear next step, with no inducement language. A guide gets at most six sections, with further points grouped as subsections under the closest one, so no point is left with a one-paragraph section of its own.

Nearly every article gets a table: key facts, costs, steps or the options compared. Its numbers come only from the article's research, your own facts or a cited source. In a comparison or a list, every fact about each option comes from a page Verand opened and verified, the official site first, and anything it couldn't verify reads “Not disclosed”. When your own product is one of the options, a disclosure line is added for you and it's judged on the same criteria as every other option. On a YMYL site, such as a financial, legal or medical one, a list keeps it out of the numbered picks and the picks table and gives it its own “Where [your firm] fits” section; it's never ranked first or called the best. Only on other sites does a list put it first.

Your compliance pack decides how. On a site the YMYL rules don't cover, a rental community under fair-housing rules included, the article names your product where it genuinely answers a section, in a sentence or two drawn only from your Brand Hub products and knowledge base, in at most one section in three. Most regulated packs allow one plain sentence, in one section, saying you offer the service that section explains, with no results, testimonials, rankings, prices or invitation to call, apply or invest. Reg D, real estate syndication and health insurance allow none in the body: the firm may appear only in the closing call to action. A site with several packs gets the strictest.

Four of the 55 each belong to one regulator's pack: Reg D, HIPAA, the SEC Marketing Rule and AICPA. Each runs only on a site its pack applies to, so a run counts 51 checks plus the ones for its own packs. Willowdale Equity carries the Reg D pack, so its runs count 52. A check with nothing to judge on a draft still runs and counts as clean, so it never dents a score. The validation engine is the same 55 everywhere.

Only when the live search results for that keyword show a featured snippet, either flagged on the brief by the research routine or seen during the run itself. Then the first pass leads with the query phrased as a question and one self-contained answer of 40 to 55 words that Google can lift. No snippet in the results, no block.

It already cannot. After drafting, every internal link is checked against the pages Verand has actually crawled on your site; an invented one is unwrapped and its sentence left intact. Then up to four real links from the brief are woven in, and links in the opening paragraph or pointing at the article's own URL are removed deterministically, because a prompt rule is a request and this one is a guarantee.