Puts a real expert's name on the article, checks every link before it ships, starts with long-tail terms, and adds three page-two queries after a month.
/plugin marketplace add mkhalid1/locul-skills
Then /plugin to install SEO, which includes this skill.
We have not measured this one. It is published untested, and the fair starting point is that a capable model already writes a long, well organised article when asked.
The skill earns its place on the steps around the draft. It requires a named author who actually knows the subject, with a real photo, a bio and links to the profiles they already keep up: LinkedIn, a personal blog, Medium, YouTube. It cites Google's own people-first guidance for why, which asks in writing whether a byline leads to more information about the author, and Google's Article markup docs, which accept author url or sameAs for telling one author apart from another. Every external link has to pass a load-and-claim check before it ships: the page loads without a login, and the sentence it supports is actually on it. Keywords go in a fixed order, long-tail and low difficulty first, medium terms only after two or three weeks of indexing. After roughly 28 days the skill reads the page's own queries, keeps those at positions 11 to 50 with buying intent and real impressions, takes the top three and works them into the page, then looks again two to four weeks later.
It is overkill for a one-off page on a site with years of authority, where the keyword order matters little. It also cannot help a team with nobody who has done the thing being written about. The author block is only as good as the person behind it.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Tier A for the checkable parts, Tier B for the prose. Material: four invented article briefs, each with a planted source list containing one dead URL, one login-walled page and one live page that does not say what the brief claims, plus a fabricated 28-day Search Console export per page with planted position 11 to 50 queries of mixed intent. Objective spine: whether the draft cites none of the three planted bad sources, whether the opening two sentences answer the query, whether the author block carries a name, photo and at least two working profile links, and whether the refresh step picks the three planted commercial, high-impression queries over the informational ones. Prose quality would need blind pairwise preference against a control.
Two parts of this skill can be checked mechanically and one cannot. The source check and the refresh selection both have right answers if the test material plants them: a dead link, a login wall, a page that does not back the claim, and a Search Console export with known commercial queries hidden among informational ones. The opening answer and the author block can be scored too.
Whether the article is more helpful is a preference question and would need blind pairwise judging. A control model told to cite sources will probably catch the dead link on its own. It is less likely to notice a live page that does not say what the sentence claims, and that check is where we would expect a real difference to show.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.Google's guidance on people-first content is short and worth reading in full. Most of the authorship rules here come from it, and the source moves when Google edits it.
For the refresh loop, a Search Console Performance report filtered to one page, sorted by impressions, with position set between 11 and 50, gives the raw list in a few minutes. A careful person with that list and an hour will make good choices about which three queries to add. The skill mainly saves the steps of judging intent and checking for overlap with other pages.
Content scoring tools such as Surfer or Clearscope compare a draft's terms and headings with the pages that rank. Pick one of them when the job is tuning many drafts against competitors. They do not check that a cited page says what you claim, and they do not choose an author.
Using the model with no skill at all is a reasonable choice for a single article on a mature site. The gains here come mostly on young sites and in the month after publishing.
--- name: helpful-article-writer description: Writes a deep, genuinely helpful SEO and GEO article for one chosen keyword, with a named expert author block (real photo, bio, linked profiles), external sources that pass a load-and-claim check, an opening that answers the query in two sentences, and copyable artifacts for the reader. It also sequences a site's keywords from long-tail upward, and runs a refresh loop about 28 days after publishing that works three commercial, high-impression position 11 to 50 queries from Search Console or Bing Webmaster Tools into the page. This skill should be used when a keyword has been chosen and an article is about to be drafted, when a new site is deciding which keywords to write first, or when a published page has a month of search data and is due for a refresh. --- # Helpful article writer ## The claim this skill is built on A page ranks and gets cited by AI answer engines when it is the most useful thing a reader can land on for that query, and when it is obvious who wrote it and why they would know. Most SEO drafting gets the first half partly right and skips the second half. It produces a long, tidy article with a generic byline, links to sources nobody opened, and keywords picked by volume. Then nobody touches the page again. Three things go wrong with the obvious approach. The author is a brand or a placeholder, so neither readers nor quality raters can tell whether the advice comes from experience. The sources are guessed from memory, so a share of them are dead, paywalled or say something else. And the keyword plan starts with the biggest terms, which a young site cannot rank for, so the first month of work earns nothing and there is no data to learn from. This skill fixes those in order: pick the right keyword for the site's stage, name the right author, write a page that answers first and goes deep after, check every source, then come back with real search data and improve the page that is already winning. ## Phase one. Choose the keyword for the site's stage Order matters here because a page can only be refreshed once it has impressions, and it only gets impressions if it can rank somewhere in the top 50. **Weeks 0 to 3: long-tail, low difficulty.** Phrases of four words or more, specific enough that the results page is thin or dated. On the 0 to 100 difficulty scales used by the common keyword tools, treat roughly 0 to 15 as the starting band for a site with few links. These are heuristics, not thresholds any search engine publishes, and each tool computes its score differently, so compare within one tool only. **After 2 to 3 weeks: medium terms.** Once the first pages are indexed and showing impressions, move to shorter phrases in the 15 to 35 band that sit above the long-tail pages in the same cluster. The long-tail pages then link up to them, and the medium page has a reason to exist. **Head terms wait for evidence.** Do not target them until several pages in the cluster hold positions in the top 20. Decision rule for each candidate keyword: - **Difficulty in the starting band and the results page shows articles:** write it now. - **Difficulty in the medium band and the site has under three weeks of indexed pages:** park it with a date, and write a long-tail child of it instead. - **Results page dominated by tools, templates or forum threads:** this is a format question, not a depth question. Hand it to a SERP-first format check before writing anything. - **You cannot tell**, because the tool shows no difficulty score or the volume is zero but the query clearly exists: treat it as long-tail and write it if a real reader would want the page. Zero-volume queries often turn out to have impressions once a page exists. ## Phase two. Name the author who knows most Pick the person on the team with the most first-hand experience of the topic. On a small company that is often the founder. On a larger one it may be an engineer, a support lead or an outside practitioner who agreed to put their name on it. The author block carries: - Full name and role, written as a specialist in this topic, for example "Builds time-tracking integrations for agencies; has migrated 40 teams off spreadsheets." Lead with the niche and a number, not a company title. - A real photograph of that person. Never initials in a circle, a stock face or an illustration. - Two to four sentences of background tied to this article's subject. - Links to every profile the person actively keeps up: LinkedIn, personal blog, Medium, YouTube, X, GitHub, a podcast. Leave out any profile with no activity in the last year; a dead channel undercuts the point. - A link to an author page on the site that collects their articles. Why this matters, stated correctly: E-E-A-T is a framework from Google's Search Quality Rater Guidelines, not a scored ranking input, and Google's own SEO Starter Guide says so. The author block is still worth building because Google's people-first content guidance asks, in its own words, "Is it self-evident to your visitors who authored your content?" and "Do bylines lead to further information about the author or authors involved, giving background about them and the areas they write about?" (checked October 2026). AI answer engines and human readers ask the same question. Markup: in Article structured data, give the author a `@type` of `Person`, a `name`, and a `url` or `sameAs`. Google's Article documentation says it understands both `url` and `sameAs` when telling authors apart, and recommends valid URLs in them. Put the profile URLs from the author block in `sameAs`, so the visible links and the markup match. **If nobody on the team has real experience of the topic**, say so in the brief and stop. Either find a practitioner willing to review and sign the piece, or pick a different keyword. Do not invent a persona. ## Phase three. Write the page: answer first, then go deep **The opening two sentences answer the query.** If someone searched "how long to keep payroll records", the first sentence gives the period and the second gives the main exception. AI answer engines lift short passages, and a reader who gets the answer immediately trusts the rest of the page more. Background, definitions and history come after, if at all. **Depth means covering what the reader will do next.** List the questions from the People Also Ask box and related searches, plus the questions people ask in forum threads about the topic. Answer each under its own heading. Stop when every one is answered. There is no word count target. **Give the reader something to copy.** Every article carries at least one artifact the reader can lift straight into their work: - a template, checklist or table they can paste into a document; - a prompt, formula or code block in a fenced block; - exact settings with the values to enter. **Link to the exact place, not the home page.** When a step happens in a settings screen, link the deep URL of that screen, for example the billing settings page of the tool, and name the menu path in words too in case the URL changes. **Write it so it does not read as filler.** Readers do not mind who or what drafted a page; they mind padding. Cut these patterns on sight: - an introduction that restates the title or opens with how the world is changing; - "X is not just Y, it is Z" contrasts; - lists of three abstract nouns used for rhythm; - a closing section that repeats the article; - hedges such as "it is important to note"; - examples with no numbers, names or settings in them. Replace each with a specific: a number, a setting, a quoted error message, a step. ## Phase four. Bring in outside sources, and check every one Videos, forum threads and long-form posts help when they give the reader something the page cannot: a screen recording of a fiddly setup, a thread where practitioners argue about a trade-off, a detailed case study. They are filler when they repeat what the page already says. Where each type earns a place: - **YouTube:** for any step that is clearer watched than read. Link to the exact moment with a timestamp (`?t=95` on a watch URL). If you embed, use a click-to-load facade or the privacy-enhanced `youtube-nocookie.com` player, so the iframe does not slow the page's first paint. - **Reddit and other forums:** for real objections and edge cases. Link the specific comment permalink, not the thread, when one reply holds the point. Quote at most a sentence and attribute it to the username. - **Medium and personal blogs:** for a deeper case study or a contrary view. Check whether the post is marked member-only; if it is, the reader may hit a paywall, so prefer an open source that says the same thing. **The load-and-claim check.** Before a link ships, confirm all four: 1. It returns the page, not a redirect to a home page, a 404 or a login screen, when opened in a private window. 2. The sentence it supports is actually stated on that page. Search the page for the specific number or phrase. 3. For video, the claim is at or within a minute of the timestamp you linked. 4. It is the most direct source. A blog post summarising a study loses to the study. Decision rule per source: - **Passes all four:** keep it, with anchor text that says what the reader will find. - **Loads but does not say what the sentence claims:** rewrite the sentence to match the source, or drop the link. Never keep both as they are. - **Paywalled or login-walled:** replace it with an open source; if none exists, keep it and label it as requiring an account. - **You cannot tell**, because the page renders by script and you cannot read its content, or the video has no transcript: leave the link out and flag it for a human to check. An unchecked link is worse than no link. ## Phase five. Publish where the first traffic is Submit the new URL to Bing through IndexNow, which Bing and several other engines accept and Google does not, and request indexing in Search Console for Google. Some sites wait weeks for Google while Bing ranks a deep page within days. That traffic is real, it also feeds the answers in Microsoft Copilot, and it is the first data the refresh loop can use. Judge a new page on both engines, not on Google alone. ## Phase six. The refresh loop, about a month later Wait until the page has a full 28 days of data. Search Console data lags by about two days, so start on day 30 or later. Read Search Console and Bing Webmaster Tools for search, and GA4 for what visitors did. A connected search-data tool that exposes all three to the agent saves the exports. 1. **Pick pages.** Sort the site's pages by clicks over the last 28 days. Work on the top ones. A page that already earns clicks has proven its topic; adding to it compounds. 2. **Pull its queries.** For each page, list every query with average position between 11 and 50. These are pages two to five: the engine already associates the page with them. 3. **Filter for commercial intent.** Keep queries containing comparison or buying words: best, vs, alternative, review, pricing, tool, software, template, for a named role or industry. Drop pure definitions. 4. **Rank by impressions.** Among what remains, take the three with the most impressions. Ignore queries with under about 50 impressions in 28 days; the position figure is too noisy there. 5. **Check for overlap.** If another page on the site ranks better for the query, it belongs there, not here. 6. **Work them in naturally.** Each query gets a real answer: a new subsection, a row in a comparison table, a question in the FAQ, a worked example. The query wording appears once in a heading or the first sentence of the new part. No repeated phrasing. 7. **Record the change.** Note the date and what changed. Update the visible modified date only if the content changed substantially. 8. **Re-check in 2 to 4 weeks.** Compare same-length windows before and after, for those three queries and the page's clicks. Do not judge on seven days. Decision rule for each candidate query: - **Position 11 to 20, commercial, high impressions:** add a full subsection. This is the closest win. - **Position 21 to 50, commercial, high impressions:** add a focused paragraph or table row, and an internal link from a related page using the query as anchor text. - **Commercial but a different intent from the page**, such as a pricing comparison on a how-to page: it needs its own page. Add it to the keyword list instead. - **You cannot tell the intent**, or impressions are too low: leave it for the next cycle and re-read it with another month of data. ## Worked example, compressed Site: a young time-tracking app for freelancers, eight weeks old. Keyword list sorted by difficulty. First article targets "how to track billable hours in google sheets", difficulty 9, results page mostly dated how-to posts. Write decision. Author: the product lead, who ran a design studio for six years and billed by the hour. Real photo, LinkedIn, a personal blog with two posts on pricing, a YouTube channel with three screen recordings. Article markup lists the four profile URLs in `sameAs`. Opening: "Track billable hours in Google Sheets with one row per task and a duration column calculated as end time minus start time, formatted as [h]:mm. Multiply by your hourly rate in a third column and sum by client with SUMIF." The artifact is a copyable sheet layout and the exact formulas. Sources: one YouTube walkthrough linked at the minute where it sets up the duration format, one forum comment on rounding to 15-minute blocks. A third candidate, a blog post cited for "most freelancers lose 10% of billable time", loaded fine but never stated that figure. The sentence was cut. Bing ranks the page on page one within eight days; Google sits at position 30 to 40. Day 32 refresh. Queries at 11 to 50 for this page include "best time tracking spreadsheet template" (position 14, 1,900 impressions), "google sheets timesheet vs app" (position 22, 1,100), "billable hours template for designers" (position 37, 640), and "what are billable hours" (position 18, 2,400). The last is informational and dropped despite having the most impressions. The first three are commercial and none ranks better on another page. Changes: a subsection comparing the template with an app, a downloadable template section with a designers' variant, and a table row. Re-check on day 56. **Verdict: refresh shipped, three queries added, next review booked.** ## Failure modes **The faceless byline.** "Written by the team" with a logo. Readers cannot tell whether the advice comes from someone who has done it, and the page gives AI answer engines nobody to attribute. **The decorative profile list.** Every social icon linked, half of them dormant since 2022. A reader who clicks finds an empty channel and trusts the page less. **Citation drift.** The link loads, the topic matches, but the specific number in the sentence appears nowhere on the page. It reads as sourced and is not, and this is the failure a quick check most often misses. **Embed bloat.** Four full YouTube iframes loaded on page open. Each pulls hundreds of kilobytes of script and the page's largest contentful paint slips, for videos most readers never play. **Chasing medium terms in week one.** The site's first ten pages target phrases in the 30s. None reaches the top 50, Search Console shows almost nothing, and after a month there is no data to refresh from. **Refresh stuffing.** The three new queries are pasted into existing sentences and headings without new substance. The page reads worse, and the queries do not move because nothing new answers them. **Promoting the informational query.** The refresh picks the query with the most impressions without checking intent, and the page bends toward a definition that brings visits with no buying interest. **Judging on seven days.** The re-check runs a week after the edit, sees a dip from normal variation, and the changes get reverted before the engine has re-crawled the page. **Google-only reporting.** The page is written off as a failure on Google while Bing sends it steady clicks nobody looked at. ## What this skill does not do - It does not fetch search data or verify links on its own. It needs an export or a connected tool for Search Console, Bing Webmaster Tools and GA4, and a way to open each source page. - It does not decide which format the results page rewards. If the top results are tools or templates, run a SERP-first format check first. - It cannot supply expertise the team does not have. With no real author, it recommends a different keyword or an outside reviewer. - It does not monitor links after publishing. Sources move and paywalls appear; a scheduled crawler catches that. - Its difficulty bands and impression floors are working rules, not published thresholds, and a site with real authority should start higher. - It does not build backlinks or plan a site's internal link structure beyond the one link per refresh query described above.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.