You don't have a linking problem. You have a deciding problem.
Once a site passes a few hundred URLs, almost any two pages can be connected somehow. The hard part is knowing which pages to link to, which pages should do the linking, and what you skip this quarter. This guide gives you a repeatable way to make that call: a scoring model for internal link target prioritization that ends in a ranked queue you can defend to your boss.
Here's the good news. You already have the raw material. Let's turn it into an order of operations.
Step 1: Name the pages you actually want to strengthen
Start with destinations, not opportunities. If you start by hunting for link spots, the pages with the most obvious openings end up dictating your strategy. That's backwards.
Build a target inventory instead. One row per important URL, with a few columns:
- URL and page title
- Page type: product, pricing, category, pillar, guide, comparison, template, docs
- Primary topic and subtopic
- Funnel stage: awareness, consideration, or decision
- The primary query or set of queries it serves
- Its business role: educate, capture demand, support evaluation, convert, retain
- The outcome you want: visibility, qualified visits, assisted conversion, direct conversion, AI citation
- Whether it is canonical, indexable, and meant to receive search traffic at all
- Its current internal link count, if you can get it
- Whatever performance and visibility data you already have
That's it. No scoring yet. You are just writing down what each page is for.
How to tell it's done. Every page in the inventory has a declared job and a declared reason it might deserve support. A pricing page is no longer sitting in the same bucket as a broad educational article. Pages that were never meant to rank are excluded from the list entirely.
Where people go wrong. They define "important" as "highest traffic." Traffic tells you which pages can help. It does not tell you which pages need help. Those are two different lists, and mixing them is the single most common reason teams prioritize internal links badly.
Pro tip: Split your library into "pages that can help" and "pages that need help" before you do anything else. High-traffic, backlink-rich pages are usually sources. New, underperforming, commercially important, or AI-invisible pages are usually targets.
Step 2: Group your library by topic, intent, and funnel stage
Now give your pages a shape. A flat list of 400 URLs is unworkable. A map of 25 topics with pages sitting under each one is something a human can reason about.
Label every page with a topic, a subtopic, an intent, and a funnel stage. Then mark:
- Pillar or hub pages
- Supporting cluster pages
- Decision pages: product, pricing, comparison
- Awareness pages answering broad questions
- Pages with no clear cluster, or with two pages competing for the same job
The test that matters is not "does this page contain the target keyword?" It is "would a reader of this page reasonably need the other one?" Those give different answers more often than you'd expect.
Hub and spoke is a useful pattern when it reflects the actual subject. The pillar orients the reader and routes them to depth. The spokes go deep and link back up when that genuinely helps. Related spokes can link sideways when the relationship is real. What you should not do is impose a fixed count. One widely used rule of thumb puts 10 to 20 subtopics under a pillar, and even that source warns against treating the number as rigid.
This is the step where internal linking hundreds of pages stops being a spreadsheet chore and starts being a map.
How to tell it's done. Every important target has a topic and intent label. Its likely supporting pages are visible in the same view. You can explain, out loud, the path from a broad question to a deeper answer to a decision page.
Where people go wrong. They treat keyword similarity as intent equivalence. Two pages can use nearly the same words and serve completely different needs. Two pages with no keyword overlap can belong in the same cluster. The other trap is linking every page in a cluster to every other page, which tells a search engine nothing because it says everything is equally related.
If you'd rather not build this map by hand, this is the part DeepSmith's Content Map does for you. It crawls your site, classifies every page onto one topic taxonomy and a funnel stage, and maps your competitors onto that same taxonomy so comparisons are like for like. Sitemaps are re-checked every 24 hours, so new pages fold in without a re-import. You still make the calls. You just make them against a current map instead of a stale export.
Step 3: Score your target pages
Here's where it gets concrete. Score every candidate destination from 0 to 5 on five dimensions, then weight them.
Target Priority = 30% business value + 25% visibility opportunity + 20% strategic or AEO importance + 15% topical centrality + 10% freshness or momentum
Before you write that down as gospel: these weights are a starting rubric, not a search engine formula and not a validated benchmark. They are here so your team argues about the weights once instead of arguing about every page forever. Calibrate them to your goals and keep the raw component scores, because the components are what make a decision explainable later.
Business value, 0 to 5. Zero is no identifiable business role. Three is a meaningful demand-generation or evaluation role. Five is a strategically critical destination with a clear revenue relationship. Use your company's real priorities here, not invented revenue estimates. A page can score high because leadership has designated it as important, even when the conversion data is thin.
Visibility opportunity, 0 to 5. This is about room to improve. Zero means no relevant demand. Three means meaningful impressions, or a page sitting close to useful visibility. Five means high-value demand with clear underperformance. Search Console gives you clicks, impressions, click-through rate, and average position, broken out by page and by query. Use the page data to score the target and the query data to check that the destination actually matches the intent behind those queries. Treat very recent data as provisional when the report says it's preliminary.
Resist the urge to invent a universal cutoff. "Positions 8 to 20" is a useful segment on some sites and meaningless on others. It depends on your query classes, your countries, your devices, and what the page is worth to you.
Strategic or AI-search importance, 0 to 5. Zero is no tracked prompt or visibility goal. Three is relevant to a tracked buyer question or a known gap. Five is the page you intend to be the authoritative source for a core buyer question, or one with a documented AI citation gap. If AI visibility is not a goal for you yet, move this 20% into business value and visibility opportunity and move on.
This is the other place DeepSmith does real work. AI Visibility tracks the prompts you care about and reports mention rate, citation rate, share of voice, and page-level citation attribution, so you can see which of your pages AI engines actually cite and which prompts you are losing. That's an input to target selection, and a good one. It is not proof that adding a link will produce a citation.
Topical centrality, 0 to 5. Zero is isolated. Three is a useful subtopic. Five is the clearest destination for several closely related reader needs. Score genuine usefulness, not your wish that everything pointed at your pillar.
Freshness or momentum, 0 to 5. Zero is obsolete or superseded. Three is current and maintained. Five is time-sensitive or newly important, where early discovery matters. This is a prioritization signal only. A publication date does not improve a ranking on its own.
How to tell it's done. You can sort your targets by weighted score, and when you look at the top 20, they make sense. Then read the raw components. A high business score with no topical fit should not win by default, and a low-traffic page can absolutely deserve the top slot if it's a strategic decision page with a real source set behind it.
Where people go wrong. Presenting the weights as a universal SEO law. They aren't. Say so in the doc, and your framework survives its first disagreement.
Step 4: Score your source pages
A great target still needs a page willing to link to it. Score your candidate sources the same way, 0 to 5 per dimension.
Source Strength = 35% external authority or link strength + 25% organic reach + 20% topical proximity + 10% engagement evidence + 10% accessibility and freshness
External authority. Pages with stronger backlink profiles are your best distributors. Google confirms links are used as signals and that PageRank remains one of its core ranking systems. Third-party scores estimate relative strength, and they are proxies, not the real thing. A crawler metric like Screaming Frog's Link Score is a useful internal-architecture diagnostic on its own 0 to 100 scale, and its own documentation says it gets harder to interpret as a site grows. It is not Google's PageRank. Don't let anyone in the room call it that.
Organic reach. Pages already pulling relevant search traffic are strong sources when the link fits the passage naturally. High traffic is not authority. It is reach, which is still worth something.
Topical proximity. The source has to be close enough that the target is a natural next step. A powerful page on an unrelated subject is not a good source. It's just a powerful page.
Engagement evidence. Use analytics to spot pages that attract qualified visits and support real journeys. Don't treat raw pageviews as proof of quality.
Accessibility and freshness. The source must be indexable and able to expose a crawlable link. Prefer current pages with a natural place to put one. Exclude anything with technical blockers or content nobody maintains.
How to tell it's done. Every one of your top targets has at least two or three plausible sources behind it. If a high-priority target has none, that's a content gap, not a linking task.
Where people go wrong. Linking from the strongest page on the site to the most important page on the site, whether or not they belong together. Authority helps only when the link belongs in the source's context.
Step 5: Score the pair, not the page
This is the step most teams skip, and it's the one that makes the whole framework work.
Ranking targets and sources separately, then connecting the top of one list to the top of the other, produces links no reader wants. Score the pair instead.
Opportunity Score = 35% target priority + 25% source strength + 25% contextual fit + 15% expected user-path value
Target priority comes from Step 3. Source strength comes from Step 4. Contextual fit, 0 to 5, asks whether the source passage and the target satisfy adjacent needs. Expected user-path value, 0 to 5, asks whether a reader who clicks actually reaches a useful next step.
Then add the column that saves you from your own spreadsheet: a plain-language explanation field. "This source should link to this target because..." If you can't finish that sentence naturally, reject the pair. It does not matter what the number says.
Here's what a good pair looks like. A high-traffic educational guide is your source. A newer comparison page is your target: important consideration-stage question, meaningful impressions, weak visibility. The guide explains how to evaluate solutions, and the comparison page is genuinely the reader's next step. That pair scores well and reads well. If the guide only mentions an unrelated concept in passing, contextual fit fails and you move on, no matter how strong both pages are.
This is also the discipline that answers which pages to link to without a meeting. The pair either has a reader reason or it doesn't.
How to tell it's done. Every approved pair has a reader reason written next to it in words, not numbers.
Where people go wrong. Treating the score as the decision. The score ranks candidates. The sentence approves them.
Step 6: Apply your gates, caps, and tie-breakers
Scores rank. Gates disqualify. Run the gates first.
Reject or hold an opportunity when:
- The target isn't the canonical page for that intent, or another page is clearly the better destination
- The source and target aren't genuinely related
- The target is blocked by noindex, a redirect, robots restrictions, or anything else that undermines the outcome
- The link would need unnatural or keyword-stuffed anchor text
- The proposed passage already has a dense cluster of links
- The destination is obsolete, thin, duplicative, or not meant for readers
- The journey it creates would confuse someone
- Two of your pages target the same query and you haven't picked the winner yet
That last one deserves a moment. When multiple pages chase the same keyword or near-identical variations, choose your canonical strategic destination before you distribute a single link. Internal links can clarify a preference. They cannot fix a page whose purpose was never decided.
Then the caps. These are workflow rules, not search engine requirements:
- Start with 3 to 5 high-confidence contextual opportunities per source article
- Don't set a universal maximum on links per page. Google says plainly there is no magical ideal number, so relevance and reader usefulness decide whether one more belongs
- Keep important pages roughly three to four clicks from the homepage, as an architecture goal rather than a ranking promise
- Give every important page at least one internal link from somewhere
- Cap how many opportunities a single target can absorb, so one page doesn't eat the whole quarter
When two opportunities score the same, break the tie in this order: clearer reader benefit, then stronger topical proximity, then higher business value, then the stronger source, then the bigger documented visibility gap, then whichever one strengthens a weak cluster, then whichever needs less editorial compromise.
How to tell it's done. Your queue is short enough to actually execute this month, and you can explain any rejection in one sentence.
Where people go wrong. Adopting a universal quota because it's easy to measure. Too few links leave real relationships unrepresented. Too many make the page noisy and the signal mushy. Start small and expand only where the reader benefit is obvious. This is the discipline that separates internal linking strategy at scale from bulk link insertion, and it's how you prioritize internal links without a quota doing your thinking.
Step 7: Point authority along the reader's journey
You now have a ranked queue. Where should the strength go?
Use your power pages as sources. Pages with strong backlinks and pages with substantial relevant organic traffic are your distributors. Point them toward newer or underperforming content when the relationship is natural.
Use business-critical pages as targets. Products, services, cornerstone guides, pricing, comparisons. But the target has to match the source reader's next need. A product page is not the automatic destination for every article you've ever written.
Follow the funnel, loosely. Awareness question to explanatory cluster page. Explanatory page to consideration guide or comparison. Consideration page to product, pricing, or demo. Decision page to proof or implementation detail when an evaluator needs it. Some readers move sideways or back up to the hub, and that's fine. This is a journey model, not a one-way street.
Use cornerstone relationships on purpose. Supporting articles link back to the pillar when the pillar genuinely helps a reader orient. The pillar links out to the deep dives most worth reading. Related pages link laterally when the relationship is clear to a human.
Don't over-concentrate. A page with the most inbound links is not automatically the right destination. Watch for one commercial URL quietly absorbing links that should be supporting other topics and other stages. Spread support across the portfolio according to your priorities and your evidence.
How to tell it's done. Your approved links support both discovery and movement. A reader following them gets somewhere useful, and your strategic pages are no longer starved.
Where people go wrong. Sending everything to pricing. It feels commercially smart and it reads as desperate.
Anchor text carries more weight than people give it. Google's guidance is to write anchor text that's descriptive, reasonably concise, and relevant to both pages, with surrounding words that supply context. Vary the wording naturally. Don't reuse identical anchors for different destinations, and don't chain links so tightly together that a reader can't tell them apart. For a link on an image, descriptive alt text does the same job.
One technical note, because it quietly breaks things: the link has to be a real anchor element with an href that resolves to a real address. JavaScript can insert one, as long as what it produces is crawlable markup. Links that exist only as script events may never be extracted at all.
Step 8: Measure the target portfolio, not the link count
Counting links you shipped tells you about your workflow. It tells you nothing about whether the strategy worked.
Measure the targets. For each one, track:
- Clicks, impressions, click-through rate, and average position
- The queries associated with it
- Organic entrances and qualified engagement
- Assisted conversions or conversion rate where you have them
- Inbound internal link count and how diverse the sources are
- Whether it moved out of orphan or weakly connected status
- Mention rate, citation rate, share of voice, and page-level citation attribution, if AI visibility is a stated goal
- Whether competitor pages are still winning your tracked prompts
Then compare by segment: by topic, by funnel stage, by page type, by source cohort. When you are internal linking hundreds of pages, the segment view is the only one that stays readable. Segment comparison is what turns a quarter of linking work into a calibration signal for your weights.
How to tell it's done. You can show a before and after per target cohort, and you can say which weights you're changing next quarter and why.
Where people go wrong. Attributing every movement to internal links. Search Console data varies by aggregation, recent data can be preliminary, and results shift with time, place, device, and history. Be the person in the room who says "this is directional," and your framework will still be trusted in six months.
A last honest note on AI. Internal links support crawlability, findability, and context, and Google's own guidance says ordinary SEO fundamentals still apply to its AI features while being clear that eligibility never guarantees serving. So track citation outcomes separately, and never promise them.
What to do next
Don't try to score 400 pages this week. You'll stall, and stalling costs more than starting small.
Pick one topic cluster. Build the target inventory for it. Score the targets, score the sources, score maybe fifteen pairs, and approve the five you can write a real sentence for. Ship those. Watch the target metrics for a few weeks. Then adjust your weights and take the next cluster.
That's the whole loop. One cluster at a time, and in a quarter you have an internal linking strategy at scale instead of a backlog. Internal link target prioritization stops being an argument and starts being a queue.
If the manual version of this is what's been stopping you, DeepSmith is built to carry the heavy parts. Content Map gives you the shared topic and funnel classification from Step 2 and keeps it current. AI Visibility gives you the citation and prompt data behind Step 3 and Step 8. And the writing pipeline places internal links during article creation, using your enriched sitemap, instead of leaving them as an hour of cross-referencing after the draft is done. You keep the judgment. You hand over the cross-referencing.
Start a free trial and score your first cluster this week.
