linkforge

Four thousand pages that never link to the product

GoodSpace grew 20× on content alone. But the page that converts — their free ATS resume checker — has five backlinks, and none of their own pages point at it. This finds the links that should already be there.

6,163backlinks to the homepage

5backlinks to /premium/ats

0from any blog post

What the gap costs

Keyword	Searches / mo	Position
ats resume	18,100	47
ats cv checker	12,100	29
ats resume checker	12,100	26
resume score	9,900	22
mock interviews	8,100	15

Commercial keywords they already have the right page for, sitting on page 3–5.

The run

A replay of a real crawl — the URLs, paragraph counts and insertions below are what the CLI actually produced against https://goodspace.ai. Nothing is fetched from your browser.

node src/cli.js --config targets.goodspace.json --limit 74

sitemap0urls in scope

fetched0pages

paragraphs0extracted

scored0above threshold

insertions0proposed

How a suggestion is made

sitemap.xml → filter by path, honour robots.txt
            → fetch sequentially, 1 request / 1.2s
            → extract <p> text + byte offsets
            → drop boilerplate: text repeated across 3+ pages is chrome
            → TF-IDF cosine, paragraph vs each target's keywords
            → apply rules, pick the least-used anchor phrasing
plan.json     exact offsets, original → replacement

The rules are the engineering

1 per target, per pageRepeating a link down one page is a spam signal, not emphasis.

3 links per page, maxPast that it stops reading as writing.

8 uses per anchor stringForces variation. Exact-match repetition is the pattern that gets sites discounted.

Never invents copyThe anchor is always existing text. The sentence is unchanged.

Structural exclusionsNo headings, existing links, code, nav, tables or forms.

Proposes onlyOutput is a patch. Nothing is applied, nothing is written to the crawled site.

What this doesn't claim. Hit rate varies by section — guide pages match at more than twice the rate of blog posts, because guides are already about the tools. A site-wide number shouldn't be extrapolated from 74 pages. There's no performance data here either, so nothing is asserted about their stack.