Generate Internal Link Suggestions from Verified URLs

AI can suggest plausible internal links that lead to pages you do not have. Prevent that by giving it an approved URL inventory and requiring every suggestion to reference an existing identifier.

Prepare the destination inventory

Include URL, page title, short purpose, status and relevant exclusions. Remove destinations that should not receive new links, such as obsolete or inappropriate pages. Keep the inventory current enough for the task.

Use a stable row ID so the model can select a destination without inventing a path. A script can then resolve the ID to the exact verified URL.

Ask for contextual links

Do not set a universal number of links per article. Add a link when it helps the reader navigate or verify something relevant.

Worked example: a migration guide

Imagine a hypothetical paragraph discussing supported import formats. A link to a verified format reference may help the reader check their file. A link to a broad sales page simply because it contains the word “import” may not.

Review the destination's actual content. Similar titles can hide different audiences or product versions.

Validate before applying

Check that each destination remains available and appropriate, that the anchor accurately describes it and that the source sentence still reads naturally. Reject self-links and unnecessary duplicates where they add no value.

For ordinary crawlable web links, Google’s guidance recommends standard anchor elements with an href pointing to a resolvable URL. Review the rendered implementation, not just the editor's preview.

Apply and inspect a small batch

Start with a manageable set and verify the resulting HTML and user journey. Keep a change log so unwanted additions can be reversed. Do not let a suggestion workflow publish automatically until its permissions and quality controls have been deliberately designed.

Measure successful implementation and reader usefulness first. Later search changes can be monitored, but a before-and-after movement does not isolate the effect of the links.

A good internal linking workflow reduces navigation friction. The model's role is to propose relevant connections inside a verified information architecture, not to manufacture more links for their own sake.

Why the inventory has to come first

Ask a model to "suggest internal links for this article" with no inventory and it will produce links. Some will point at pages that exist. Some will point at pages that sound like they should exist on your site, with paths assembled from the article's own vocabulary. A reviewer cannot tell the two apart by reading; every suggestion has to be checked, which removes most of the time saving.

Supplying an inventory changes the task from generation to selection. The model chooses among rows it was given, by ID, and a script turns the ID into the verified URL. Fabrication becomes impossible at the point where it matters, and the reviewer's job shrinks to judging relevance, which is the part that needs a person anyway.

The inventory is also where editorial policy lives. Pages that should not receive new links, such as retired guides, legal pages, campaign landing pages with an end date, or pages for a different audience, are simply not in the file. A rule in a prompt can be ignored; a row that is absent cannot be selected.

A prompt shape that produces reviewable suggestions

The output should be a table the reviewer can accept or reject row by row, with enough context to decide without opening the source article each time:

You are given (A) the article as numbered paragraphs and (B) an inventory of approved
destination pages: id, title, one-line purpose, audience.
Suggest internal links only where the reader of a specific paragraph would benefit
from the destination as a next step or a verification.
Return: paragraph_number, quoted_sentence, destination_id, proposed_anchor_text,
reader_benefit (one sentence), confidence (high/medium/low), reason_if_low.
Rules: select destination_id only from (B); never write a URL; at most one link per
paragraph; skip paragraphs where no destination clearly helps; do not suggest a link
already present; anchor text must describe the destination, not the source.

The quoted_sentence field lets a script verify that the sentence exists unchanged, which catches the occasional paraphrase. The confidence and reason_if_low fields let the reviewer start with the uncertain rows, which is where the wrong suggestions concentrate.

Review criteria that keep links useful

Accepting a suggestion is an editorial decision, and it helps to make the criteria explicit so that different reviewers apply the same standard:

Apply the accepted rows with a script that inserts standard anchor elements pointing at the resolved URLs, and inspect the rendered HTML, not the editor preview, on a small batch before running more. Keep the batch's change log so any link can be reversed. Then measure what the links are for: whether readers follow them, and whether support questions about "where do I find X" decrease. Ranking movement afterwards is worth noting and impossible to attribute, which is fine, because navigation was the goal.

Put this into practice

Copy the worksheet columns below into a spreadsheet and keep one row per item you check. The filled row is an illustrative example, not a reported customer result; replace it with your own verified records.

Use the following prompt only after supplying the records it requests:

Suggest links from this article using only approved destination IDs. Return source passage, destination ID, anchor and reader benefit. Omit weak matches. Never invent URLs or change the supplied destination meaning.

Research context

Content automation can propose useful links when destinations and reader relevance are verified. The related Ahrefs starting points are 7 Ways to Automate Content Marketing with Agent A and Automated SEO: What It Is and How It Works in 2026. This guide’s checklist, examples and proposed workflow are independently written; they are not results of a SEOVision experiment.

Continue with the next task

• Cluster Keywords with AI and Catch the Wrong Groupings
• SEO Prompt Engineering: Reusable Briefs, Guardrails, and QA
• Technical SEO Audit Checklist: A Practical 30-Minute Workflow

Sources

• 7 Ways to Automate Content Marketing with Agent A — Research starting point; not an endorsement of this original workflow
• Automated SEO: What It Is and How It Works in 2026 — Research starting point; not an endorsement of this original workflow
• Google: crawlable links — Primary documentation for the stated platform behavior