Two numbers per page, and what each combination means
Two numbers per URL: how many pages link to it, and how many it links out to. Read as a pair they describe four kinds of page, and knowing which one you are looking at decides what the fix is.
The short answer
Every page has an inbound count and an outbound count, and the pair sorts your site into four quadrants: hubs that link out widely, authorities that many pages point at, connectors that do both, and isolated pages that do neither. The isolated quadrant is where the work is, and the counts alone say which fix applies.
Two numbers, and the four pages they describe
Inbound and outbound are usually reported as separate columns and read separately, which loses most of the information. The combination is what identifies a page.
One caution about the counts before any of this is actionable. An inbound number that includes your navigation is a number that tells you nothing, because every page will be linked from every page. The figure worth reading counts links from the body of a document, which means the report has to separate content from chrome, and that separation depends on your markup saying which is which.
What to do about each quadrant
- High inbound, low outbound: a terminus. Everything points here and nothing leaves. Common on pricing, contact and product pages, and often correct: a checkout page should not be sending people away. It becomes a problem when the page is editorial, because a well-supported article that links to nothing is a page whose support goes no further.
- High inbound, high outbound: a hub. The healthy shape for a category page, a pillar page or a good index. Nothing to do except notice which of your pages are actually hubs, because the list rarely matches the one in your head.
- Low inbound, low outbound: nearly invisible. One link in, one link out, no route through it. These are the pages that quietly stop existing, and the fix is a decision: link it properly from somewhere relevant, merge it into a page that has an audience, or remove it.
- Low inbound, high outbound: a connector nobody uses. Tag archives and generated indexes live here. The page distributes attention across dozens of destinations and receives almost none itself, which means the distribution is thin. If nobody needs the page, it is a candidate for
noindexrather than for more links.
The pattern to look for is not a single page in the wrong quadrant. It is a whole template sitting in one, which tells you the problem is structural: every tag archive in the bottom right, every product page in the top left.
Sort by inbound ascending and read the first two screens.
That single operation is worth more than any of the analysis above. The pages at the top of that list are the ones your site barely acknowledges, and among them there is almost always something you spent real time on. Fixing those is a morning of adding links from pages that are genuinely related.
The very bottom of that sort, inbound zero, is a separate check with a higher rating, because a page with no route is a different problem from a page with a thin one.
An inbound count is a fact about other pages
How many pages link to this one is a property of every other page on the site. You cannot read it off the page being counted, in the same way you cannot tell how many people are talking about you by examining yourself.
This one needs the full crawl
An inbound count is a fact about every other page on the site, so it cannot be read from the page being counted.
The instant search on this site audits a single page, so rather than show you a verdict it cannot support, this guide sends you to the place the check actually runs.
The outbound half of the pair is visible from one fetch. It is the half that matters least, which is a reasonable summary of why this check needs the crawl.
Does this affect AI search?
Only as a proxy, and it is worth being precise because link counts get treated as a currency. Nothing about retrieval counts your internal links. What the numbers correlate with is discovery and context: a page linked from several relevant places is a page more likely to be found, and a page whose neighbours are about the same subject arrives with that context attached.
The quadrant that matters most here is the bottom left. A page with one link in and one out is reachable only if something happens to follow that exact link, and an agent working through two or three hops from a starting point has no particular reason to. Thin connection is close to invisible for a visitor with a small budget, even where it is adequate for a comprehensive crawler.
The outbound side has a separate and honest use: links out of a page are how a model reading it learns what else you have written on the subject. That is not about counts, it is about whether the links are in the prose and describe their destinations, which is the anchor text check.
Why a table of counts is rated at the bottom
Because the counts are not defects. There is no threshold for inbound links, no correct number of outbound ones, and no state in this table that removes a page from anything. What you have is a description of your site, which is what a Refinement means: information that improves a working site.
Every genuine defect adjacent to this check is rated higher and reported separately. Zero inbound links is an orphan. A route that stops is a dead end. A page nobody can reach in fewer than eight clicks is a depth finding. This check is what those three are computed from, and on its own it is a lens.
Which is not the same as being useless. The sort in the previous section is one of the highest-value ten minutes available in an audit, precisely because it is a description: it shows you what your own site thinks is important, and that is frequently news.
Counting both directions with a shell
Outbound is one fetch. Inbound needs every page, which is why it is the slow half.
TARGET=/guides/carbon-steel
# outbound: internal links leaving the page
curl -s "https://example.com$TARGET" \
| grep -o 'href="/[^"#?]*"' | sed 's/href="//;s/"$//' | sort -u | wc -l
# inbound: how many other pages link to it, and which
curl -s https://example.com/sitemap.xml \
| grep -o '<loc>[^<]*</loc>' | sed 's/<[^>]*>//g' \
| while read -r page; do
[ "$page" = "https://example.com$TARGET" ] && continue
curl -s "$page" | grep -q "href=\"$TARGET\"" && echo "$page"
done | tee /tmp/inbound.txt | wc -l
# how many of those are the navigation rather than the content?
wc -l < /tmp/inbound.txtIf the inbound count comes back equal to your total page count, the link is in a template and the real editorial figure is zero. That distinction is the difference between a page that is genuinely supported and one that merely appears in a menu, and it is the most common way these numbers get misread.
Questions this check raises
- What does a high outbound and low inbound count mean?
- It is a hub nobody links to: a page that does a good job of routing readers onward while receiving almost no authority to pass along. Category indexes and resource lists end up here constantly. The fix is inbound links from the pages that benefit from it, not more outbound links.
- Is a high inbound count always good?
- No. A page with hundreds of inbound links that are all site-wide footer links has one link repeated, not hundreds of endorsements. The number is only meaningful once you separate navigation from editorial links, which is why the count is a starting point rather than a verdict.
- How many internal links should a page have?
- There is no target, and any number quoted as one is invented. What matters is that the links are relevant and that no page has zero. A page with three well-chosen links to genuinely related content is in better shape than one with sixty from a template.