Why your link profile matters more now that AI engines cite sources

We see the same pattern across several B2B clients. They rank second or third in Google for their most important term. Ask ChatGPT or Perplexity the same question and three other names come up: a trade publication, a comparison site, and a competitor with less traffic. The follow-up question is always the same: how do we get in there?

The answer has less to do with your own website than you would hope. It has more to do with the rest of the web.

Being cited is not the same as ranking

A classic search engine gives you ten results and leaves the choice to the user. An AI engine gives one answer and names a handful of sources. That selection is stricter. The engine retrieves documents, weighs them, and picks the sources that best support the answer. What matters is not only whether your page answers the question, but also whether your claim is confirmed elsewhere.

—An engine that gives a single answer has no ten blue links to hide behind. So it picks sources it sees confirmed somewhere else.

That is the core of GEO, generative engine optimization. You are not optimizing for a position, but for the chance that a model uses your page as evidence. And that chance is closely tied to how often, and in what context, others refer to you.

Why links now weigh more, not less

For years the expectation was that links would matter less. With link building for AI search engines, the opposite tends to happen, for a few practical reasons:

• Crawlers still find most new pages through links. Without a reference you enter the index late, or not at all.
• Mentions without a link count too. Models connect a brand name to a topic, even when there is no hyperlink attached.
• A claim that appears on several independent domains is easier for a model to confirm than a claim that only lives on your own site.
• Retrieval indexes and training data lean on the same public sources: trade media, Wikipedia, industry bodies, forums, knowledge bases.
• Consistency compounds. If your brand name keeps appearing next to the same theme, you build an association that ad budget does not buy.

Put briefly: your own site is your evidence, the surrounding web is your reference.

A citable link profile looks different

The classic SEO link profile was judged on domain authority, anchor text and volume. Citations depend on other things. In practice we look at:

• Topical relevance over authority score. A trade publication with limited reach that covers your niche weekly often weighs more here than a general news platform.
• Sources the engines demonstrably cite. Look at which sites already appear in the answers for your topics. That is your target list, not an abstract list of high-authority domains.
• Context around the link. A sentence explaining what you do and for whom is more usable to a model than a name in a link list at the bottom of a page.
• Repeated mentions over time. One article is a moment. Four mentions in a year in the same place is a pattern.
• Consistent facts. If your proposition, location or approach is described differently in five places, you give the model reason to doubt.

Note that all of this is simply good PR. That is no coincidence. On this point, the distance between brand building and findability is shrinking.

Five steps for a team of two

With a small team and a limited budget you can work on this perfectly well. Not through volume, but through direction.

1. 1Build a fixed set of twenty to thirty prompts around your main themes, buying questions and comparisons. Run them through two or three engines and record which sources get cited.
2. 2Pull out the five to ten domains that keep coming back. That is your target list for the next six months.
3. 3Produce one thing per quarter worth citing: your own numbers from your client base, a benchmark, a named method, a checklist that goes further than the rest.
4. 4Ask for the mention actively. Respond to articles with a correction or addition, offer an expert quote, contribute a guest piece. Trade editors are chronically short on substantive input.
5. 5Make sure your own page supports the claim properly: a clear date, a named author, a source for every figure, and one unambiguous phrasing that others can copy.

This costs attention rather than money. Budget half a day a month for measuring and roughly three days a quarter for making. That fits alongside the rest of the work.

What we do not know yet

Let us be honest: nobody knows the weighting. The providers do not publish how their source selection works, and that selection changes. The same prompt can name three sources today and four different ones next week. So do not build a business case on a citation count.

We also do not know how large the commercial effect is. In most accounts we see, traffic arriving through AI engines is still modest, though the intent is often higher than with organic traffic. Whether that holds depends on how fast behaviour shifts. What is certain: investing in relevant mentions pays off without AI as well, through referral traffic, brand recognition and classic SEO. That makes it a defensible choice, even if the assumptions about engines look different a year from now.

Measuring without a reliable dashboard

The tooling for this topic is young. Part of the measuring is still manual work. What works in practice:

• Run your prompt set monthly on the same day and record the results in a single sheet. You are watching the trend, not the individual measurement.
• Track mentions of your brand name, including those without a link, using a simple alert or monitoring tool.
• Separate referral traffic from AI engines as its own channel in your analytics, even while the numbers are low.
• Ask every new lead how they found you. That one line in your form often tells you more than your attribution model.

At REX.digital we treat this as part of the foundation rather than as a standalone campaign: first establish where you are mentioned today and where you are not, then pick a target list, then publish and pitch, and recalibrate each quarter based on what the engines actually cite. Start small. Set up your prompt set this week and measure your baseline. Without that baseline, in six months you will not know whether anything changed.