Link Building For AI Search Is A Domain-Level Bet

Ask a link builder what a good placement looks like and you will get a number. DR 50, maybe DR 55, on a site with real traffic, ideally with the link pointing at the page you are trying to move. That answer has been stable for a decade because it was tuned against Google's ranking behaviour, where a mid-authority editorial link on the right page is a sensible unit of work. It is now being used, without revision, to buy visibility in AI answers. Nobody re-derived it for the new surface. They just kept buying the same thing and filed it under a new heading on the invoice.

The DR Band Everyone Buys Is Not The Band That Gets Cited

Start with the two datasets side by side, because the gap between them is the whole argument. Ahrefs took the top 1,000 pages ChatGPT cited in September 2025, pulled through Brand Radar and then run through batch analysis. Among the cited pages that rank in search, 65.3% sit on domains rated DR 81 or higher. The median Domain Rating is 90. That is not a mid-authority distribution with a long tail at the top. That is a top-of-market distribution with a modest tail below it, and the 11.7% of cited pages sitting at DR 0 to 20 is the exception that people quote precisely because it is an exception.

Now the spending side. Reporter Outreach surveyed 500 SEO professionals in the first quarter of 2026, a sample that skews toward agency owners and specialists with three or more years in the job. Fifty-three percent of them target the DR 40 to 60 band. Nine percent go above DR 70. Reporter Outreach sells outreach services, so read the framing as interested, but the distribution is consistent with every link-buying conversation we have sat in: the mid band is where price and plausibility meet, and it has been the default for years.

Put plainly: the band that absorbs most of the industry's link spend is not the band that shows up in the citation data, and the band that does show up is the one almost nobody is buying. That is an uncomfortable finding rather than a damning one, because there is a perfectly good reason for the mid-band default. DR 81+ placements are not a line item you add to a retainer. They are earned through actual news value, original data or a genuine relationship with a desk, and they do not come with a rate card. The industry did not choose DR 40 to 60 because it was optimal. It chose it because it is purchasable.

Which matters, because purchasability is now pushing prices in the wrong direction. Seventy-six percent of that same survey will pay $300 or more per link, 31% land in the $500 to $1,000 range, and 64% spend at least $3,000 a month. Fifty-eight percent increased spend year on year, and 75% expect prices to keep rising. Rising prices in a band with a weak relationship to the outcome you are now buying for is the definition of a budget that needs re-examining, not a budget that needs growing.

Link Building For AI Search Runs At The Domain, Not The Page

The second number in the Ahrefs set is the one that should change how a programme is built, and it has had almost no attention. Of those top 1,000 cited pages, 67.3% carry a URL Rating between 0 and 10. The median UR is 6. URL Rating is Ahrefs' page-level authority score, driven by links pointing at that specific URL. So the pages ChatGPT cites most sit on extremely strong domains and have almost no links of their own.

Read that twice, because it inverts the standard brief. The standard brief is: decide which page you want visible, then buy links pointing at that page. On this evidence the page-level signal is close to irrelevant to citation selection, while the domain-level signal dominates it. The unit of the bet is the domain. The page is just where the answer happened to be found. A link pointing at your pricing page is buying page authority on a surface that does not appear to weigh page authority, and the domain lift you get as a side effect is the part that was actually working.

The content mix in the same study supports the same reading. Wikipedia accounts for 29.7% of those cited pages and homepages or landing pages for 23.8%, neither of which is a category you win by pointing links at a URL. More telling: 28.3% of the cited pages have no organic keyword visibility at all. They are not ranking. They are being retrieved and cited off the strength of the domain they live on, which is exactly what you would expect if domain-level trust is doing the selection work.

There is a caveat worth stating before anyone reorganises a quarter around it. Ahrefs sells the tools that produce DR and UR, and a study showing that their domain-level metric predicts citation membership is not a disinterested finding. The sample is also one engine at one point in time, which is a narrow base for a structural claim. What makes it credible is that the direction is consistent with what the other datasets say, and that the UR result actively works against the commercial interest, because it tells buyers that one of the things link vendors sell has less effect than assumed.

Low Domain Rating Still Earns Citations, On Some Engines

The obvious objection is that all of this reads like an argument for giving up, and that low-authority placements demonstrably do earn citations. That objection is correct, and there is now good data behind it. Stacker analysed 895 stories distributed between 6 May and 11 September 2026, which generated roughly 212,000 pickups across its network, and tracked citations across six surfaces over the four weeks after distribution. Domains under DR 40 accounted for a fifth of the citations the network earned. Low authority is not a wall.

But the blended figure hides the finding that should govern your budget. Split by engine and the tiers stop behaving alike. Claude cited sub-DR40 domains at 1.8% and DR 41 to 60 at 2.6%. AI Overviews sat at 1.8% for both tiers. ChatGPT cited sub-DR40 domains at 0.05% and the middle tier at 0.04%. The spread between the friendliest engine and the least friendly one is more than fiftyfold at the bottom tier, which means the question 'does low DR work' has no general answer. It has an answer per engine, and they disagree violently.

One more detail from the same study deserves a line of its own. Citations plateaued at around 90 pickups for domains under DR 40. Past that point, additional syndication stopped adding citations. Volume substitutes for authority up to a ceiling and then stops substituting, which is the most useful shape anyone has put on that trade-off, and it gives you a spend limit rather than a vague warning. Stacker sells syndicated distribution, so the incentive runs toward showing that its network earns citations. The plateau finding cuts against that incentive, which is a decent reason to trust it.

Note also that these citation rates are small in absolute terms across the board. A 2.6% rate is the top of this table, not a benchmark to be disappointed by, and anyone quoting AI citation rates in double digits is measuring something narrower than they are describing. The comparison that matters here is between tiers and between engines, not against an imagined baseline.

What This Does To A Link Building For AI Search Budget

Three changes fall out of reading the three datasets together, and none of them require a bigger budget. The first is that the engine you care about has to be named before the placement target is set. We built a decision table for weighing AI citation channels on the assumption that the channel was the variable. On this evidence the engine is a variable of equal weight, and a DR 45 placement that is a reasonable buy for Claude and AI Overviews visibility is close to worthless for ChatGPT. Same placement, same price, two different answers depending on a question most briefs never ask.

The second is that the domain, not the URL, is the thing being bought. That reframes what counts as a win. A homepage mention on a DR 88 publication has historically been graded below a deep link with exact anchor into a commercial page, and on the AI side that grading is backwards. It also means the existing programme is probably underperforming its own reporting, because the placements that moved domain-level trust were being scored on anchor and target-page criteria that the citation data does not reward.

The third is the one that costs money to act on and saves more money than it costs. A mid-band programme running at $3,000 to $6,000 a month, which is the largest single spend segment in the survey at 26% of respondents, buys roughly six to ten placements at prevailing rates. Those same dollars concentrated into one or two genuine top-tier earned placements per quarter buy far fewer links and land in the band where the citation data actually concentrates. That is a real trade with a real risk, because top-tier placements cannot be guaranteed and mid-band ones effectively can. We ran a version of this argument with 500 link builders and 897 journalists on opposite sides of the same pitch, and the journalist-side data is where the risk in it shows up: top-tier desks reject most of what reaches them, so the concentration play only works if the underlying story is genuinely fundable.

For B2B SaaS teams the calculation is usually cleaner than it looks, because the buying research is concentrated in ChatGPT and Google's answer surfaces rather than spread evenly across six engines. When we took Zenity from zero to citations across four AI engines, the placements that moved the needle were not the ones with the best anchors. They were the ones on domains the engines already trusted to be about that subject.

Repoint The Programme Before The Next Invoice

This is an afternoon of work, not a strategy offsite, and most of it is re-scoring things you have already paid for.

Worth separating two things that get conflated whenever this argument comes up. None of it says mid-band link building is dead, and anyone reading it that way is looking for permission to stop doing work that still pays. Mid-band editorial links still move Google rankings, still drive referral traffic, and on two of the six surfaces in the Stacker data they perform at least as well as anything else. The claim is narrower and more awkward: the case for the mid band rests on the Google side of the ledger, and the AI side that is increasingly used to justify the budget does not support it. If the programme is sold on citation visibility, it should be bought against citation evidence. If it is sold on rankings, say that instead, and the existing brief is fine. Our own link building engagements get split along that line at the scoping stage now, because a single number covering both surfaces hides the trade rather than resolving it.

Do this next. Open the last four link invoices and add one column recording the domain rating of each placement and one recording the engine you were trying to influence when you bought it. Most teams cannot fill the second column, and that is the finding. A programme that cannot say which surface a placement was for has been buying a mid-band average against a distribution that is not average anywhere, and the prices in that band are forecast to rise for the next two years. The cheapest moment to repoint it is before the next renewal, not after the next report.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.