Authority, links and brand in AI search

A link vendor will show you the half of this research they have picked. Here you will be able to read both halves side by side and thus judge the pitch yourself.

The studies that say links matter

SE Ranking looked at 129,000 domains, and they started looking at 216,524 pages as well¹.

Then there's a separate analysis of 1,000 AI Overviews and 4,200 URLs. In that one, domain authority at +0.61 was the strongest citation factor².

The studies that say they do not

AirOps built their study with Kevin Indig and traced all 353,799 pages from 16,851 queries as ChatGPT retrieved them. Their conclusion was that backlinks and domain authority show no positive correlation with citation. Backlinks and citation are slightly negatively correlated when it comes to ChatGPT³.

The next bit is their larger run of 43,233 queries. And yes, sites in the middle of the authority range got cited on average more often than the strongest domains⁴.

Cyrus Shepard says something in between which also sounds right to me. His review scores the AI citation factors proposed by 55 separate studies. A lot of the studies reviewed find a domain authority effect, but are often finding it weak. Domain authority was in the middle of his list and its score was a 5.0 out of 10⁵.

What actually explains the contradiction

Both of the camps are sound and their disagreement flows from where each one looked. Much of the detail is on the how AI answers are built page. First you have to get a page into the set that feeds the answer. Most of that retrieval is driven by pretty ordinary rankings anyways.

1. Earn good links, and you'll build good rankings. That is uncontroversial to both camps.
2. The rankings themselves decide which pages to pull for a question, and how to order them.
3. Once a page is in that retrieval set, the quality of the answer matters. That means you have to put the answer near the top, with headings that match the question and facts that are specific.

So those whole-web studies were finding a link effect because the rankings of those pages were built on links. It's not a question of a link effect in the AirOps case, because the current retrieval set already reflects the current rankings of every page in it.

The slight inverse correlation in the data AirOps published is exactly what you'd expect from this setup. Those weaker-linked pages are winning on relevance instead of links, which is what the selection step rewards.

Brand works differently, and moves on a different clock

And brand is a whole other type of authority that has little to do with anything I described. That brand authority is what Dan Petrovic calls primary bias, which is mostly a sense the model forms in training to rate a brand's relevance to a topic.

If you are already tied to a subject and are familiar to the model from training, then you're more likely to be able to be selected when you turn up in the results. His estimate of three to six months for this to change is based on the pace of model retraining, not the pace of a campaign⁶.

The 6.8 Shepard review score for brand and entity trust is higher than the 5.0 Shepard review score for domain authority⁵. So, the soft signals are rated better than the paid ones.

This is probably the most urgent topic for ChatGPT visibility.

No amount of page editing, or schema markup or content from your blog can change that.

The brand thresholds, and why a local firm should ignore them

This is where I question what people actually repeat from this research. SE Ranking also looked at two brand thresholds¹.

• Roughly 240,000 brand searches, linked to AI Mode citation odds about a third better
• Over 1.16 million visitors, three times as likely to be chosen as a source as sites under 2,700

A single-office firm is able to take its own niche and town and be unambiguous to the engines.

• Same firm name everywhere
• Same address and phone
• Same practice descriptions
• Same attorney names, across your site, your bar listing, the legal directories and any press

The shift away from brand scale is called entity consistency. Any good entity cleanup can be finished in an afternoon.

Brand searches are surprisingly useful for ordinary Google rankings as well.

The route in that is not links and not brand

And this one is a little surprising. Even though 31.0% of the AI Overview citations in Ahrefs' study didn't rank in Google's top 100 for the query, they were still able to turn up as sources. For instance in that non-ranking group a full 18.2% of citations are YouTube URLs⁹.

Let's say, a small firm will get cited on questions its website can't reach because one attorney will put up a video answering them, for example. That video is using a real title and transcript, and no link building at all. In local legal markets this rarely if ever happens. The off-site page in this guide is about that work.

Where authority sits in local search, for comparison

But local search offers a useful comparison, because the factors point the same way. In the 2026 Whitespark survey, where 47 practitioners scored 187 map pack factors, none of the top 24 factors are link factors. The first link factor, 25th on the list of factors. The top 24 include the primary category, the business title, the hours, the rating, and the review count¹².

For a local firm it often makes little sense to buy links ahead of free profile work that ranks higher.

What Google asks for, which a firm can genuinely produce

Nothing in Google's guidance on content quality is an authority score. Google asks, does the content even show first-hand expertise, or does it just hide who wrote it? Each time you publish a page, ask yourself whether the page cites its sources or whether you would trust it as a reader and client. For topics that affect money, health or safety, which legal advice does, these signals get even more weight¹¹.

Attorney credentials are easier to show than most signals and seem to be skipped nowadays.

• Named attorneys on the pages they wrote, with bar admissions, jurisdictions and the courts they appear in
• Reviewed-by lines where an attorney checked a page they did not write
• Dates that reflect real revisions
• A press page listing coverage with links
• Bar roles and speaking history, with links that prove it

Plus, that list doesn't need an outreach campaign or a budget.

What I would do with an authority budget

1. A first week on consistency. Having to match your name, address, phone, and practice descriptions across your site, your Business Profile, your bar listing, Avvo, Justia and Martindale is tedious, slow, and practically free. Any mismatch left in those listings may weaken the entity that everything else attaches to.
2. Authority you have already earned. You've likely been named on the pages of the bar committees and the law school alumni lists, among others. Ask them, and the local papers that mentioned you, to link to you because you're already named on the page. In practice, most are happy to add it.
3. Expertise signals on the pages. Author bylines with admissions, reviewer lines, honest revision dates. This is the part that Google describes wanting¹¹.
4. Twenty recorded video answers. The videos and their transcripts are not dependent on your site's authority at all⁹.
5. Paid link work, and only after the rest. If you do buy links, ranking movement on named queries is due every three months. But never judge it on citations or visibility or domain rating.
6. Brand work, judged in quarters. None of this brand work is going to show up for you in a month⁶. Likewise, I'm very sceptical when I read about agencies who lifted a firm's AI visibility in a weekend or moved its brand in one week without a model retrain.