Best AI Prompts for Backlink Analysis (2026)

Prompts for backlink analysis#

Act as an SEO analyst. Below is a backlink export for my competitor [PASTE COMPETITOR DOMAIN] with columns [LIST YOUR EXPORT COLUMNS, e.g. referring domain, target URL, anchor text, link type, first seen]. Using only this data, list the 20 referring domains most worth pursuing, the page type each links to, and why. Do not add any domain or metric that is not in the export; write "not in data" instead. [PASTE YOUR EXPORT]

Act as a link-building strategist. Compare my referring domains [PASTE MY EXPORT] with the referring domains of three competitors [PASTE THREE EXPORTS]. Find domains that link to at least two competitors but not to me, group them by site type (publisher, directory, resource page, community), and give one reason per group why a link might be attainable. Mark every claim that depends on information outside the exports.

Act as a technical SEO. Here is an anchor text column from a competitor's backlink export [PASTE ANCHORS WITH COUNTS]. Group the anchors into branded, URL, generic, exact-match and descriptive phrases, show each group's share of the total, and flag anything that looks unusually repetitive. Say what the pattern might mean and what it cannot prove.

Act as a digital PR lead. From this list of pages that link to competitors [PASTE URLS, TITLES AND LINK CONTEXT] pick the resource pages, guest contribution pages and roundups. For each, draft a short personal outreach angle based on what I can offer [DESCRIBE YOUR CONTENT, DATA OR EXPERTISE]. No templates sent in bulk, no claim that I have a relationship I do not have, and no offer of payment for a link.

Act as a risk reviewer. Here are links pointing to my site from my own export [PASTE REFERRING DOMAINS, ANCHORS, LINK TYPES AND DATES]. Flag patterns that Google's link spam policies describe, such as paid or exchanged links, sitewide footer links and keyword-heavy anchors. For each flag, quote the rows that triggered it and say what a human should check before any removal or disavow request.

What to add to any backlink analysis prompt#

• The raw export itself, with its column names, wrapped in tags such as <export> so the model can tell data from instructions. Anthropic and OpenAI both recommend XML tags to separate reference material.
• The data first and the question last. Anthropic advises putting long inputs above the query, and Google's Gemini guide says to place instructions at the end of the prompt.
• A rule to answer only from the data, and to write "not in data" otherwise. Anthropic suggests asking the model to quote the relevant parts of documents first, which makes unsupported claims easier to spot.
• The output format you want, such as a table with named columns. Google's Gemini guide notes the model picks its own format if you do not specify one.
• One or two example rows of the output you expect, kept in the same structure. Gemini's guide recommends varied, consistently formatted examples.
• What you can honestly offer a site (data, a tool, expertise), so outreach angles are real. This is our own advice rather than vendor guidance.

A worked example#

Using the fourth prompt, filled in:

Act as a digital PR lead. From this list of pages that link to competitors, pick the resource pages, guest contribution pages and roundups. For each, draft a short personal outreach angle based on what I can offer. No templates sent in bulk, no claim that I have a relationship I do not have, and no offer of payment for a link.

What I can offer: a free open dataset of UK electric-vehicle charger locations, updated monthly, with a CSV download and a methodology page. I run a small EV news site and can write a 600-word expert piece.

<pages>
1. https://example-evguide.co.uk/resources | "Useful EV resources" | list of tools, links to competitor A's charger map
2. https://example-motoringblog.co.uk/write-for-us | "Write for us" | guest posts accepted, links to competitor B
3. https://example-greenweekly.co.uk/roundup-oct | "Green news roundup" | links to competitor A's report
</pages>

A good result:

• Sorts each page into resource page, guest page or roundup, using only the rows pasted.
• Gives a different angle per page that ties to the dataset or the article offer, not one template with names swapped.
• Says nothing about the site owner that is not in the pasted rows, such as names or past contact.
• Leaves out any offer of payment, free product or reciprocal link.
• Marks anything it cannot know, such as whether the page is still maintained, as "not in data".

The full prompt library is on our prompt library page.

Limits of this guide#

Google's documents tell you which link practices break its policies and how to mark paid links. They do not say that any prompt wording improves rankings or that an AI tool finds better link targets, and we did not test that. We also could not confirm how any particular SEO tool scores authority or spam, so we say nothing about those metrics. The prompts assume you can export data from a commercial or free link tool.