Links or presence? What separates the sites AI Overviews quote

Hypotheses and checks

Part 1 tested position. This part started with links, then moved to organic visibility and coverage across the market.

1. Ranking positions account for the citation gap. A 10x spread remained after accounting for positions in part 1. Position mattered, but didn’t account for the whole gap.
2. Link profiles help explain the remaining gap. My starting hypothesis. I compared referring domains, then repeated the count with followed links. Neither produced the separation I saw on traffic and keywords.
3. Cited sites have broader organic visibility. I checked estimated traffic, ranking keywords and top-3 keywords. The selected cited groups were ahead on all 3. That describes the groups; it doesn’t yet explain the picking.
4. Broader query coverage accompanies a higher citation rate per appearance. The broadest-coverage group beat the narrowest on citations per appearance, too. The bands in between didn’t rise steadily.

I also checked whether a site had to rank in the top 20 of the citing search. 119 sites were cited without it, so that isn’t a requirement.

How I measured this

Every domain that appeared in the sampled results or their AI Overviews got an Ahrefs profile. The analysis uses 3 groups:

I define the middle group more strictly than “absent from the citing search”. These 119 sites held no top-20 position on any of the 6,984 searches that carried an answer.

The profile comparisons use 300 quoted-and-ranking sites, all 119 quoted-without-ranking sites, and 300 never-quoted sites. I selected the quoted groups by citation count and the never-quoted group by number of ranking appearances, because it has no citations to rank by. The market-coverage analysis uses all 2,497 ranking domains.

That last bit is a real weakness and it belongs here rather than in a footnote: 2 of the 3 groups are ranked on the outcome I am studying, and the third is ranked on something else. I checked which way it pushes. Selecting all 3 groups by coverage instead gives a link gap of 1.73x against the 1.50x this page reports, so the selection I used understates the gap it reports rather than inflating it.

Medians throughout. A few enormous sites would pull the averages out of shape: the average monthly traffic across the quoted-and-ranking 300 is 43,481,538 visits against a median of 28,628. For links I count referring domains (RDs): distinct sites linking to a domain.

1. The link gap is the small one

Median RDs: 6,164 for quoted-and-ranking sites, 4,355 for quoted sites without rankings, 4,104 for sites that ranked and were never quoted.

The first group is 50.2% ahead of the third on RDs, and 41.5% ahead of the middle group. Set that against what the same comparison looks like on traffic and keywords.

Before that, the part I got wrong when I first wrote this up. I called 1.5x “the link gap” and put it at the top of the page, and it is a gap that depends on where I cut the groups.

The link gap runs anywhere from 1.35x to 2.72x. So does the traffic gap. Neither of them is a constant of nature, and 300 was a number I picked.

The last column is the one that holds. Traffic separates these groups between 3.1 and 3.5 times further than links do, at every cutoff from 200 down to the whole group. At the cutoff this page uses, the 2 confidence intervals do not overlap: links 1.26 to 1.90, traffic 2.88 to 8.89. That comparison is the finding. Either gap on its own is a number with a cutoff attached to it.

Median ratios: 300 quoted-and-ranking sites against 300 never-quoted ranking sites. Bars start at 1x, the point where the 2 groups match.

Site by site, the 3 groups cover the same ground.

Each dot is 1 site. The black bar on each row is that group’s median. The axis is cut at 40,000. The count on the right of each row says how many sites sit past it.

The link count separates the extremes better than the middle. In the middle, the same range holds cited and ignored sites. I couldn’t use it on its own to decide which of 2 ranking competitors the answer would pick.

A code-hosting platform makes the point at the extreme: 2,530,517 referring domains, appearances on 8 sampled searches, and no citations. At the other end, a blockchain project’s help center had 55 referring domains and collected 12 citations from 18 appearances.

A huge profile went uncited; a small one was cited repeatedly. The count doesn’t settle it. Whether the links and the page are relevant to the query is still open.

2. Counting followed links tells the same broad story

Then I counted only RDs with followed links.

Medians: 4,093, 3,102 and 2,526. Quoted-and-ranking sites are now 62.0% ahead of never-quoted sites, against 50.2% before. That’s a change, but it doesn’t turn the link gap into the traffic gap.

The axis is cut at 40,000. The count on the right of each row says how many sites sit past it.

So the all-link count wasn’t hiding a large gap behind nofollow. The distributions still overlap. I’d next check who the links come from: do sites in this niche provide a clearer distinction than total RDs? That’s a different hypothesis, and I haven’t tested it here.

3. The visibility gap is 5x

Ahrefs-estimated organic visits per month, at the median: 28,628, 18,390 and 5,550, in the same group order.

Median ranking keywords: 1,756, 2,070 and 312. On top-3 keywords, the first group is 5.66x ahead of the third.

Here are the same 3 groups compared pairwise on both measures, with the second group in each row as the baseline.

Every row moves further on traffic than on links. The middle row moves furthest: almost no difference in link count, more than 3 times the traffic.

The axis is cut at 200,000 visits. The count on the right of each row says how many sites sit past it.

The quoted-without-ranking group is the useful comparison here. Its median RDs are only 6.1% above the never-quoted group, but traffic is 231.4% higher.

Medians for the 119 sites cited without a top-20 position anywhere in the sample, against the 300 never-cited ranking sites.

Almost the same link count, a very different level of organic visibility. Traffic separates those 2 groups much more clearly than RDs do.

But 90 sites in the never-quoted group have more estimated traffic than the median quoted-and-ranking site. I’d go through those separately. My hypothesis: some are large sites about other things, ranking here on a few pages. Their size and link authority may get those pages into the top 20 without giving the site broad coverage of this market.

I haven’t classified them yet. I’d compare their main subject, niche-query coverage and ranking pages with cited sites of similar size and position.

If that holds, total traffic is partly a measure of how big the site is. What I’d want instead is the traffic and keyword footprint inside this niche. A site can be big elsewhere and barely present here.

4. How much of the market you show up in

For all 2,497 ranking domains, I counted how many of the 7,601 queries each appeared on in the top 20. That is the coverage measure here.

Share of domains never cited, by number of sampled searches where they ranked. All 2,497 ranking domains.

87.1% of the 1,190 domains appearing on exactly 1 search were never quoted. Of the 138 appearing on 100 or more, just 6 were never quoted: 4.3%.

Ranking on a couple of queries isn’t the same as being present across the market. Most of those narrow sites weren’t cited once. Almost every site in the 100+ band was.

More rankings also mean more opportunities to be cited. So I checked the citation rate per appearance: how often the site was cited on a query where it ranked.

Citation rate per ranking appearance, by number of sampled searches where the site ranked.

For sites on 1 to 4 queries, 15.9% of ranking appearances also carried a citation. For sites on 500 or more, 35.0%. More than double, per appearance.

It isn’t a steady climb. The next 2 bands dip to 15.3% and 14.2%, then the 100 to 499 band reaches 21.6%. The biggest difference is at the broadest end.

1 thing that band rate hides, and it sharpens the point rather than softening it. 15.9% is citations divided by appearances across the whole band, which is the right way to build a rate. The typical site in that band gets 0.0%: most of those 1,784 sites were never cited at all, and the 15.9% belongs to a minority of them. In the 500+ band the median site runs at 27.2% against the band’s 35.0%, and the sites inside it range from 1.8% to 81.5%.

I also checked whether a few large sites are carrying this. They are not: the single biggest domain contributes at most 6.5% of the appearances in any band. Giving each site 1 vote and comparing the narrowest band against the broadest puts the difference well past any threshold I would argue about.

I wouldn’t turn that into a promise of more citations for every extra keyword. What it supports is a broader hypothesis: sites with extensive coverage of relevant niche queries are cited more consistently than sites that intersect a few of them.

I’d next hold position and site size still, then compare niche-focused sites with general ones. Does coverage still separate them? That would tell me whether it’s adding information beyond size and rank.

5. The sites quoted without ranking

119 sites were cited while holding no top-20 position on any of the 6,984 searches that carried an answer. Their median was 4,355 RDs: just 6.1% above the never-quoted group, and below the 6,164 of quoted-and-ranking sites.

Estimated traffic was 18,390 visits a month at the median, 231.4% above the never-quoted group.

They got in without a top-20 ranking anywhere in this market and without a much bigger link count. I’d put their cited pages next to the pages that rank and get skipped. Query relevance, commercial or informational content, depth of coverage: which of those separates them?

Where this meets other studies

Ahrefs studied 75,000 brands. Its correlations with AI Overview visibility were 0.295 for referring domains and 0.218 for backlinks. My own subset is 86 sites holding 200 or more positions, and I owe you a sentence about what a correlation means at that size.

It came out at 0.21 for referring domains, with a 95% interval running from −0.00 to +0.41. At 86 sites that figure can’t be told apart from zero. I checked 8 metrics on the same 86 sites, and after correcting for the fact that I checked 8, only 2 of them hold up: Domain Rating at 0.28 and the share of donors passing a followed link at 0.27. Referring domains, followed referring domains, total keywords and estimated traffic all land in the same 0.20 neighborhood with intervals that touch zero.

So I can’t put 0.21 next to Ahrefs’ 0.295 and call it agreement or disagreement. Theirs is built on 75,000 brands and mine on 86 sites, and mine is a number I would not defend to 2 decimal places. What both sets of figures do say, at very different sample sizes, is that raw link counts are nowhere near a strong signal.

In the Ahrefs study, brand web mentions had a correlation of 0.664 and brand search volume 0.392. Those associations were stronger than the ones for link counts.

Floyi’s 2026 topical authority report gets closer to the coverage question. It mapped 29,319 queries into 42 topical maps and measured how a site’s coverage related to citations.

For AI Overviews, the coverage correlation was 0.51. Across the engines studied, Domain Rating correlations ranged from 0.09 to 0.16. Sites ranking for at least half of a map’s topics appeared in 30.8% of that map’s citing answers; sites ranking for under 5% appeared in 0.14%.

That points in the same direction as my coverage result. Floyi measured defined topics; I counted appearances on a query set. I can’t call mine topical authority yet.

What it changes about part 1

Part 1 ended on a gap I could not close. Among the 86 sites holding 200 or more positions, citation rates ran from 2.8% to 44.8%. A 10x spread survived accounting for where those positions sat.

I’d expected links to explain more of it. The counts give me much less separation than traffic and keywords do. They haven’t supplied the explanation I was looking for.

Coverage moves up my list: how much of this market the site appears in. The broadest group was more likely to be cited at all, and was cited more often per ranking appearance.

On the next SEO plan, I’d start with the site’s relevant query coverage, commercial and informational. Where is it absent? Where does it rank while the answer cites somebody else? Those are the gaps I’d investigate before setting another link target.

My working hypothesis is that broad niche coverage supports citation visibility. This capture shows an association. Whether expanding that coverage earns more citations still needs testing.

Worth going deeper

Next I’d look at the pages. 3 things are worth coming back to.

Query coverage. Are the citation leaders focused on this market across the site, or is 1 page doing the work on a site about everything? And are the 90 high-traffic sites without citations big elsewhere but narrow here? I’d map their niche queries and the pages covering them.

Content type. Product and service pages, or informational articles? I’d compare what gets cited with what ranks on the same queries and gets skipped.

Content depth. A broad treatment of the topic and its related queries, or a short, focused answer? I’d compare cited and ignored pages. Length alone won’t tell me whether the relevant questions are covered.

That is where I’d take the next part: niche coverage at site level, and relevance, type and depth at page level.

There are 3 more things I want to take apart. Who the linking domains actually are: whether a site’s donors sit inside this market or somewhere unrelated to it, and whether the cited sites are carrying links from the same places as the ignored ones. Which queries sit behind each traffic figure, and how many of them belong to this niche. And what the pages at the top of these results are made of: type, depth, and how directly they answer the query they rank for.

Each one is another round of exports against a monthly limit I pay for, and I have spent this month’s. So this is where part 2 stops. The rest goes into the next one.