9. One Search Article Worth Judging

Search can be judged for Ledgerlane with one article. The article answers a question contractors already type into Google. It links to the offer page, and you confirm that Google can index it. Then you write down a dated starting point, so that in a few months you can say whether search brought qualified leads or only visits. The offer page, the embedded booking form, Plausible and the lead sheet are already in place from the earlier chapters. This chapter adds the first channel that feeds them.

Finding, writing and checking the first search article

Start with what people actually type. Open Google in a private window, or sign out first. Autocomplete, the list of suggestions that drops down as you type, is personalised by your own history and location, and signing out reduces that bias. Type a stem such as "hmrc penalty letter" and stop. Write down every suggestion exactly as it appears, spelling mistakes included. Then add modifiers one at a time: "hmrc penalty letter what", "hmrc penalty letter sole trader", "self assessment bill", "self assessment bill higher than expected", "self assessment bill why". Autocomplete tells you that a phrase gets searched. It does not tell you how often. Google does not explain in detail how predictions are made, so treat it as a list of real phrasings, not a ranking.

Next, look at the People Also Ask box, the set of expandable questions that sometimes appears in the results. Note any questions that match your stems. Don't build a plan around it. The box comes and goes from one search to the next, and published estimates of how often it appears disagree with each other. If it shows up, take the questions. If it doesn't, you've lost nothing.

Then open Google Keyword Planner inside a Google Ads account and set the location to the United Kingdom. Two warnings matter here. The column called "competition" measures how many advertisers bid on a phrase. It says nothing about how hard it is to rank in the ordinary results. And without active ad spend, the monthly volumes may appear as broad bands rather than exact counts. Use them to compare phrases with each other, such as whether "payments on account" is bigger or smaller than "penalty letter", and nothing finer than that.

Google Search Console's Performance report is the fourth tool, and it works differently. It shows clicks, impressions, click-through rate and average position for queries that already found your pages. For a new site, that means it confirms topics after you publish. It does not discover them beforehand. It also hides rare queries in its table to protect searchers' privacy, so the rows you see won't add up to the totals. Once the article is live, filter the queries for words like "penalty" or "payments on account", sort by impressions, and group them by page.

Now choose the topic. A good one is a specific, personal problem with facts you can check on gov.uk, and it shows up in autocomplete. The pick here is the first Self Assessment bill that turns out much bigger than expected. That surprise has a mechanical cause called payments on account. These are two advance instalments toward next year's tax, each usually half of the previous year's bill, due on 31 January and 31 July. In a first Self Assessment year, you pay the full bill plus the first payment on account by 31 January. So a contractor who expected to pay one year's tax can face roughly one and a half times that in a single payment. That is the big-bill trigger in the evidence file, explained by a rule anyone can verify. A late-filing appeal would be another candidate, but Ledgerlane doesn't handle appeals, so the article would point readers to a service it doesn't sell.

You'll often see queries sorted by intent, such as informational, comparing or ready to buy. Use that as a working framework of your own, not a rule from Google. "Why is my first self assessment bill so high" is someone trying to understand. That person needs an explanation first and an offer second.

Now the brief and the draft. Open a new chat in the Ledgerlane Marketing project. Ask the assistant for an article explaining why a first Self Assessment bill is larger than expected. Tell it to use only the facts register rows for deadlines, payments on account, late-filing penalties and late payment, and the evidence-file lines about the big-bill trigger. It should tag every sentence with its row or line, or write "no source". It should promise no outcome that depends on HMRC, and end with a link to the offer page. Then check every tagged fact against the register. The register has late filing at £100 up to three months late, then £10 a day for up to ninety days, then £300 or five percent at six and twelve months. Late payment is five percent at thirty days, six months and twelve months, plus interest. Watch for the same fault the course found on the offer page, a penalty figure that has lost its condition. The register's note still applies too: sole traders inside Making Tax Digital fall under the points-based late-submission system, so check HMRC's penalty page for the tax year before quoting a stage. The payments-on-account facts aren't in the register yet. Add them as new rows with the gov.uk page as the source, the founder as owner and today's date as checked, then rerun the tags. Last, do a voice edit against the brand guide: short sentences, numbers with their conditions, no "stress-free", no reassurance in place of information.

On-page structure comes next. Write a title that describes this page and no other, such as "Why your first Self Assessment bill is bigger than you expected". Google may rewrite titles it finds vague or repetitive. Write a meta description, the short summary under the title. It informs the snippet Google shows, but you can't set the snippet directly. Don't aim for a word count, because Google says it has no preferred length. Stop when the question is answered. And never change the date on the article just to look fresh. Change it when the facts change.

Then link the article both ways. In the article, add a link to the offer page as an ordinary link element pointing at a real address, with descriptive text such as "monthly bookkeeping for one- and two-van contractors", not "click here". Google follows proper links. It may not follow a button that only works through a script. On the offer page, add a link back to the article, so every important page has at least one link pointing to it. Both links use the plain page path with no UTM tags. Those labels are for traffic arriving from outside campaigns. On an internal link they would overwrite where the visitor really came from.

After publishing, check indexing. In Search Console, paste the article's address into URL Inspection and read the status. Open the Page indexing section and check two things: that crawling and indexing are allowed, and which canonical address Google chose. If you fix anything, use Test live URL to see the current version. You can click Request indexing, but it doesn't guarantee anything. And "URL is on Google" means the page is eligible to appear, not that it will.

Two settings can quietly block a page. First, if your robots.txt file blocks the address, Googlebot never fetches the page, so it never sees a noindex tag on it. Blocking and noindexing are different jobs, and the blocking one hides the other. Second, a canonical tag, which names the preferred version of a page, is a signal that Google weighs, not a command it must obey. If Google picked a different canonical, find out why. Add the article to your sitemap as well. A sitemap is a hint that lists your canonical addresses, and Google ignores its priority and change-frequency fields, so don't spend time setting them.

Visibility in AI answers is the last thing to record, and you can only observe it. Google says AI Overviews and AI Mode need nothing beyond normal Search eligibility, and its page on AI features and your website is worth reading for yourself. Vendor studies suggest each platform cites different sources. Ahrefs found that about twelve percent of URLs cited by ChatGPT, Gemini, Copilot and Perplexity were in Google's top ten for the same prompt. Writesonic found that only about 3.8 percent of cited sources were shared by all four. These are vendors' own numbers, not independent research, but the pattern is clear enough: ranking in Google doesn't settle what an assistant cites. The method is simple. Keep a fixed list of five to ten queries taken from your autocomplete notes. Run them on a schedule, say every two weeks, in each assistant you care about. Log the answer, the cited URLs and a screenshot. Repeat each query, because answers vary from run to run.

Then write the baseline. Make a sheet dated 6 November 2026 with the query list, impressions at launch (zero or close to it), the sources each assistant cites today, and the three measures that will judge the channel: qualified leads in the lead sheet that can be traced to the article, read as a minimum because internal links carry no campaign labels, form submits, and hours spent writing and maintaining it. Whether you publish this exact article is your call. The sheet only records what is true on the day you start.

A new Search Console report for AI visibility

Search Console now has a generative AI report, which Google launched on 3 June 2026. It shows impressions from AI Overviews and AI Mode, but no clicks and no queries. Clicks from those features are still counted inside the classic Performance report, and it isn't settled whether a filter can separate them out. That has a practical consequence. If you add the new report's impressions to the classic report's totals, you count the same appearances twice.

For a small site, read it as a visibility signal. It tells you whether the article shows up in AI answers at all. To estimate visits that came from AI tools, look at referrers in Plausible instead, since that is a record you control. The smallest useful step is one new column in the baseline sheet, AI-feature impressions, filled in on the same fortnightly schedule as your query log and kept apart from the classic totals.