Nadia Mohamed
Technical SEO
How to Run an Internal Linking Audit Without Inventing the Rules
What does an internal linking audit actually tell you?
An internal linking audit tells you which of your pages the rest of your site treats as important, and whether that matches which pages you think are important. That is the whole of it: a comparison between your stated priorities and your revealed ones, expressed in links. Google is direct about why this matters: links are both a relevance signal and a discovery mechanism. An internal linking audit is how you find out what yours are currently signalling.
That framing matters because it sets the limits of the exercise. An internal linking audit does not tell you whether your content is good, whether your titles are right, or whether you deserve to rank. It tells you where link equity and crawl attention are pooling, and where they are not reaching.
It also sits inside something larger. If you are running a wider technical audit, internal linking is one workstream within it, not a substitute for it — and in the twelve-phase framework it is a phase with its own inputs and outputs rather than a standalone project.
There is a second thing worth saying before the process, because it shapes how much of the standard advice you should trust.
TL;DR — Key takeaways
- An internal linking audit compares the pages you say matter against the pages your links say matter. Where those disagree is the finding.
- The most-repeated numeric rules in this topic have no source behind them. “At least five internal links per page” and “nothing deeper than three clicks” are conventions, not evidence.
- Google’s actual published position is narrower and more useful: every page you care about needs a link from at least one other page, and beyond that Google says there is no ideal number.
- Four checks are genuinely required: orphan pages, click depth, pages carrying a single inbound link, and anchor text that describes its target.
- For AI visibility, internal linking matters because it drives indexing — Google states plainly that there is no special optimisation for AI Overviews or AI Mode.
- Prioritise findings by what the affected page is worth to the business, not by the count of issues in the crawler’s export.
- Every fix needs a verification step with a date attached, or you have not finished the audit.
The numbers everyone repeats, and where they come from
I extracted the three highest-ranking pages for this topic in full before writing this: Screaming Frog’s tutorial (DR 87), Semrush’s guide (DR 92), and Respona’s internal link audit guide (DR 76). Between them they contain no original sourced statistic at all. Semrush quotes Google’s documentation twice, which is proper practice. Beyond that, every number in all three pages is asserted.
That is not a coincidence of which three I picked. On the Ahrefs SERP overview for this query, six of the nine ranking results are published by SEO tool vendors and three are user-generated posts on LinkedIn, Reddit and Medium. The advice is largely written by companies selling the crawler you would run, which is a reason to check the reasoning rather than a reason to dismiss it.
Two assertions in particular travel everywhere, and both deserve to be handled carefully.
“Every page should have at least five internal links.” Respona states this as a baseline. Its own FAQ, further down the same page, then says there is no exact number. Google’s guidance is narrower and clearer: every page you care about should have a link from at least one other page on your site — one, not five — and Google says there is no ideal number of links a page should contain, adding only that if you think it is too many, it probably is.
Five is not wrong as a working target. It is simply not a finding, and treating it as one leads teams to manufacture links to hit a quota rather than to serve a reader.
“Nothing should be more than three clicks deep.” All three pages assert a version of this. None cites a primary source for the threshold. It is a reasonable convention — deep pages genuinely do get crawled less — but the specific number three is folklore, and an audit that fails a page purely for sitting at depth four is enforcing a rule nobody established.
I am flagging this rather than quietly repeating it because the alternative is to add a fourth unsourced page to a topic that already has plenty. Verifying a statistic before publishing it is the method behind everything on this site, and it applies most where the received wisdom is thickest.
The practical upshot: use these conventions as triage thresholds, not as pass/fail gates. A page at depth five that earns revenue needs attention. A page at depth five that nobody should ever reach is working as intended.
The four checks that are actually required
These four appear in every serious treatment of the topic, and each maps to something Google has actually published.
Orphan pages
An orphan page has no internal links pointing to it. It may still be in your sitemap and may still be indexed, but nothing on your own site vouches for it.
This is the one check with unambiguous documentation behind it. Google’s link best practices state that every page you care about should have a link from at least one other page. That is the floor, and orphans sit below it.
Find them by crawling your site and comparing the crawl’s URL set against your sitemap, your analytics, and your Search Console data. Anything present in the second group but absent from the first is orphaned. In Search Console these pages frequently surface as crawled but not indexed — that status is often an internal linking problem wearing an indexing costume.
Before you conclude a page is orphaned, confirm robots.txt is not blocking the path. A disallowed URL looks identical to an unlinked one in most crawl exports, and the fixes are completely different.
Click depth
Click depth is the number of clicks from your homepage to a page. Deeper pages are crawled less often, which slows how quickly changes are picked up.
Measure it, sort descending, and then — this is the part most checklists skip — filter the list by commercial value before you act. Depth is only a problem for pages that should be getting attention.
To confirm a deep page is genuinely under-crawled rather than merely deep, look at what your server logs show about crawl frequency. Logs turn an inference into an observation, and they will occasionally tell you a depth-six page is crawled daily, which ends the argument.
Pages carrying one internal link
A page with exactly one inbound internal link is one edit away from being orphaned. This is the check that ages best, because it catches problems before they become problems.
Sort your inlink counts ascending and look at everything with a count of one. For each, ask whether the page is worth more links or whether it should be consolidated into something else. Both are valid outcomes. Adding links to a page nobody needs is how sites end up with navigation nobody reads.
Anchor text that describes its target
Anchor text tells readers and Google what to expect at the other end. Google defines the standard as what Google calls good anchor text: descriptive, reasonably concise, and relevant both to the page it sits on and to the page it points at.
There is also a test you can apply mechanically, published by Google: Google’s own anchor text test is to read the anchor text on its own, out of context, and check whether it is specific enough to make sense by itself. “Click here” fails immediately. “Our guide” fails. “How we price technical audits” passes.
Export every internal anchor on the site, deduplicate, and sort by frequency. The generic anchors rise to the top on their own. While you are in there, check placement too — Google advises you don’t chain links together, because adjacent links are hard to tell apart and each one loses the surrounding text that gives it context.
The checks worth running second
These matter, but they are cleanup rather than structure. Run them once the four above are resolved.
Broken internal links. Straightforward, and worth fixing on sight. Update the link to the live URL rather than relying on a redirect to paper over it.
Internal redirect chains. Links pointing at URLs that redirect, sometimes more than once. These usually arrive in bulk after a restructure — if you have run a migration and never swept the internal links afterwards, assume you have them. Repoint the links at the final destination.
Links that only exist after JavaScript runs. Google only crawls real anchor elements with an href attribute. The important nuance, and the one that stops this becoming a panic: JavaScript-inserted links are fine if they render as anchors. The failure mode is not JavaScript itself, it is markup that never becomes a real anchor — a click handler on a span, or a router attribute with no href. This is common enough in client-side routers that render links without href to be worth a targeted check, and the broader picture is in links that only exist after JavaScript runs.
Deliberate exclusions. Before you “fix” an unlinked page, check whether it is excluded on purpose. The distinction between canonical and noindex signals decides whether a page is missing or correctly hidden.
Prioritising what you found by business impact
An internal linking audit will hand you a list of several hundred issues with no ordering. Most checklists stop there, which is precisely the complaint I hear most from in-house teams: the output is either too shallow to act on or too long to triage.
Order the findings by the value of the affected page, not by the severity label in the export. In practice that means three passes:
- Pages that make money or generate leads. Any structural issue here is worth fixing this week. A single orphaned product or service page outranks fifty broken links in an archive.
- Pages that support the first group. Cluster and pillar content whose job is to funnel authority. Fix issues that break that flow.
- Everything else. Batch it, schedule it, and do not let it hold up the first two.
This is also how you get engineering time. A ticket that says “fix 340 internal linking issues” gets deprioritised; a ticket that says “three service pages are orphaned and cannot be reached from navigation” gets picked up. Framing the work in terms of the metrics that justify the work to stakeholders is the difference between an audit that ships and one that sits in a spreadsheet.
Worth keeping in proportion: this topic draws around 450 searches a month in the US at a keyword difficulty of 12, according to Ahrefs Keywords Explorer. It is a routine maintenance task, not a crisis. Treat it that way in your planning.
What internal links do for AI assistants
This is where the topic attracts the most speculation, so here is what is documented.
Google states there is no special optimisation for AI Overviews or AI Mode — no additional requirements beyond ordinary SEO. Its position on extras is equally plain: no AI-specific files or markup required, schema.org structured data included.
So the mechanism is not exotic. It runs through indexing. To be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed to be cited and eligible to appear with a snippet. An orphaned page that never gets indexed is not eligible to be cited — not because AI systems dislike orphans, but because it never reaches the pool they draw from.
Google lists internal links among the fundamentals for AI features, alongside allowing crawling and keeping important content in text. That is the whole claim, and it is worth stating carefully: internal linking supports AI visibility by supporting indexing. Anyone telling you there is a distinct internal-linking strategy for AI citation is describing something no engine has documented.
Two qualifications. This is Google’s account of Google’s surfaces; it says nothing about how other assistants retrieve, and AI Mode and AI Overviews behave differently even within Google. Both surfaces may also use query fan-out, issuing several related searches behind one question — which rewards having a genuinely covered topic rather than one page trying to answer everything.
None of this contradicts the case for showing up in AI Overviews or for structured data for AI search. It qualifies it: do those things because they help users and indexing, not because an AI-specific requirement exists.
Checking whether the fix worked
An internal linking audit that ends at “we added the links” is unfinished. Every recommendation above should carry a check, and the checks need dates.
For each fixed page, record the state before you touched it: inbound internal link count, click depth, indexed status in Search Console, and impressions over the preceding 28 days. Then set a review date far enough out to be meaningful. Crawling and reprocessing can take days to months depending on how often Google decides a page needs refreshing, so a two-week review will usually show you nothing and tempt you into concluding the fix failed.
At the review date, check three things in order:
- Is the page now indexed? For a previously orphaned page this is the whole ballgame. If it is still not indexed, the internal link was not the binding constraint and you need to look elsewhere.
- Has crawl frequency changed? Your logs answer this directly.
- Have impressions moved? Last, and least reliable in isolation — impressions move for many reasons, and attributing them to one internal linking change is usually overclaiming.
Set this up once and it costs nothing to repeat. Tracking AI referral traffic covers the assistant-side instrumentation, and a dashboard that shows the before and after is what turns a one-off internal linking audit into something you can re-run each quarter without rebuilding the measurement each time.
If nothing moved and the page is indexed and crawled normally, the honest conclusion is that internal linking was not this page’s problem. That is a useful result. Record it and move on.
Frequently asked questions
What is an internal link audit?
It is a review of how the pages on your site link to one another — which pages receive links, how many, how deep they sit, and what the anchor text says. The output is a prioritised list of pages whose internal support does not match their importance.
How often should you run one?
Run an internal linking audit quarterly on most sites, plus once immediately after any migration or restructure, which is when internal linking breaks in bulk. Sites publishing several times a week benefit from checking new content monthly, since fresh pages are the ones most likely to be under-linked.
How many internal links should a page have?
At least one, from a page other than itself — that is Google’s stated floor. Beyond that, Google says there is no ideal number. The widely-quoted “five per page” is a convention with no published research behind it. Use it as a rough target if it helps you, but do not fail a page against it.
Do internal links still matter for AI search?
Yes, through indexing rather than through anything AI-specific. A page must be indexed and snippet-eligible before it can be cited in Google’s AI features, and internal links are part of how pages get discovered and indexed. Google is explicit that no additional optimisation is required for those features.
What tools do you need?
A crawler that reports inlink counts and click depth, plus Search Console for indexation status and, ideally, server logs for crawl frequency. The specific crawler matters less than running the same one consistently, so your before-and-after numbers are comparable.
Is there a real limit on links per page?
Not a published one. Google’s position is that there is no magical ideal number, with the caveat that if it feels like too many, it probably is. Judge by whether a reader can tell the links apart and whether each one earns its place.
What is the difference between an orphan page and a page that is simply deep?
An orphan has no internal links at all. A deep page has links, only a long path from the homepage. Orphans are always worth investigating; depth is only worth acting on when the page has commercial value. Checking the GEO audit checklist alongside this one will catch the cases where a page is reachable but still not being retrieved.