Engineered to get you found.

A crawler with a polite user agent reads up to fourteen pages, chosen to cover what you sell, who you sell to and how you talk. From them a model writes two documents you can edit: a business profile (offer, audience, pains, differentiators, location, competitors, the page an article should send readers to) and a brand-voice brief (tone, point of view, spelling, do and don't lists, one sentence that sounds right).

Reachability, the https redirect, one canonical host, indexability, the homepage title and description, a single H1, the viewport, structured data, social tags, the page language, a blog section, robots.txt, the sitemap with its counts and dates, titles across pages, time to first byte. Each finding carries its fix in plain words; one button copies the list for whoever runs the site. Two findings block everything else: a noindex, or a robots file that shuts Google out. The check runs again with every plan refresh.

Each seed is expanded through public autocomplete data in your language and market, then the whole set is clustered, labelled by intent (informational, commercial, transactional) and scored from 0 to 100 for how much it matters to this business: relevance to the offer, buying intent, and whether you could credibly answer it. Phrases about competitors, jobs or unrelated products are dropped.

The plan avoids topics your site already covers, mixes formats (how-to, cost guide, comparison, checklist, mistakes, buyer's guide, local guide, single question) and spreads intent roughly 60/25/15. The first four are the most searched, most answerable topics, so the first week earns trust. You can skip, restore, reorder by writing next, or add your own topics. When the plan runs out, thirty more are planned.

Title, slug, meta title (≤60 characters), meta description (140–155), the key takeaways it must deliver, an H2/H3 skeleton where every section states its purpose, four to six FAQ questions readers ask next, two to four internal links chosen from your own pages, one call to action, and a prompt for the cover image.

A live web search gathers five to eight specific, current facts or statistics relevant to the brief, each attached to its source. Only these may be cited. If the search finds nothing solid, the article is written without statistics rather than with invented ones.

The writer follows the brief in your voice: specifics over adjectives, the question answered in the first 120 words, tables where readers compare, no filler, none of the machine tells. An editor pass then cuts what still sounds generated, checks the keyword sits naturally in the first paragraph and one heading, and removes any claim the notes do not support. Health, money and legal topics run under a conservative rulebook: describe and compare, never diagnose or promise.

The drafting is done by a large language model — the same family of model behind ChatGPT and Claude — working from the brief, your voice document and the research notes, never from a blank prompt. That is why an article costs you $3 rather than the $300 a freelancer charges, and saying so is more honest than implying a room of writers. What the model is not allowed to do is the part that matters: it cannot invent a statistic, because a figure without a URL from the research pass does not survive the editor; it cannot invent a link, because every link is checked against pages we actually found; and it cannot publish, because nothing reaches your site until you approve it or you switch that off yourself. The machine writes the draft. The rules, the sources and the last word are not the machine's.

Every link in the finished article is verified: internal links must be pages we found on your site; external links must be sources from the research pass. Anything else becomes plain text before you see it. There is no link network between customers and never will be.

A generated cover (or a typographic one when image generation is off) is uploaded with alt text. The article is rendered to clean HTML with heading anchors, FAQ schema and your meta tags, and published to WordPress, Shopify, Wix, Webflow, Ghost, Sanity, Supabase or your webhook. Where nothing can be written into — Squarespace, Framer — it arrives formatted to paste, and you tell us the address. With nowhere to publish at all yet, it waits on a preview page Google is told to ignore. Until you switch on auto-publish, it waits for your approval.

With Search Console connected, we submit the sitemap the check found if Google lacks it, resubmit after every published article, and ask Google's URL inspection about the homepage and every article: indexed or not, and why. A page still out after three days becomes an item on your list with the exact link to press, because Google keeps the Request indexing button for site owners. The day Google lets the page in, the item closes itself.

After the research and then weekly, the pages missing a title or description, or carrying one too long, too short or shared with another page, get a draft written from the page text and the searches Search Console already shows it for. Each shows the current text, the new text and the reason. Approve, and the connected WordPress, Shopify or Webflow is updated and the live page fetched once to confirm. Fields a platform keeps out of its API become a note for your developer. Nothing changes without an approval.

Once Search Console has data, we look for articles of ours that Google already shows for a search just below the first results. The page is relevant enough to rank; its wording is not quite the searcher's. The model rewrites the title, the meta description, the first heading and the opening paragraph around that exact search, keeping every fact, and writes one linking sentence for a related article of yours. You approve; we republish in place and read the position again after 14 and 28 days. Put back undoes it.

We generate three questions a customer would ask when choosing a provider like you, put them to an assistant with live web search, and record whether you are recommended, at which position, and which businesses are named instead. Over time this is the scoreboard the plan plays against.