Obility Editorial · · 4 min read
Links do their work before the answer is written
Ask whether backlinks still matter for AI search and the usual reply is “indirectly”, followed by a correlation chart. That is true and not much help when you have to decide which link to chase. It is more useful to follow the order in which an answer engine works. First it searches. Then it reads the pages it found, writes the answer from their text, and cites the pages it used. Links matter a great deal in the first step and very little in the second.
The engines describe the first step themselves. Google’s documentation on AI features says that to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to appear in Google Search with a snippet, with no additional requirements, and that both features may run several related searches across subtopics. Google’s guide to its ranking systems still lists link analysis and PageRank among its core ranking systems. Microsoft’s Bing Webmaster Guidelines, which also cover Copilot, say strong internal and external linking supports discovery, authority evaluation and grounding eligibility. OpenAI’s help center says ChatGPT search sometimes rewrites a question into targeted queries and sends them to partner search providers. In every case, a link helps a page reach the pile of results the model reads.
The answer names what the retrieved page says
Once the pages are retrieved, the link graph has done its job. The model writes from the text in front of it. Suppose a buyer asks for project management tools suited to small agencies. If one retrieved page is a guide that describes your product in two plain sentences, the answer can name you. If the retrieved page links to your homepage from a sentence about something else, that link did nothing for this answer. The engine cites the page it read, and the page it read never said what you do.
So a link and a mention do two different jobs. A link to your page raises the odds that your page is retrieved for the queries it matches. A mention on someone else’s page, in a passage that answers the buyer’s question, gives the engine a second page that can name you. Plenty of link building produces the first without the second: footer credits, resource pages that list bare URLs, guest posts on unrelated topics with a link in the author bio.
Judge each link opportunity by the page it sits on
Before chasing a link, ask two questions about the page it would sit on. Would that page itself be retrieved for any question your buyers ask? And does the passage around the link say what you do, for whom, in the words a buyer would use?
Take two placements a payroll software company might be offered. One is a logo and link on a large publisher’s partners page, alongside two hundred others. The other is a paragraph in a mid-sized accounting firm’s guide to paying overseas contractors, saying the firm uses the product because it handles cross-border payments. The first may pass more ranking strength to the homepage. The second is the kind of page that gets retrieved when a buyer asks how to pay contractors abroad, and it already contains the sentence an answer would need. For AI answers the second is worth more. It stays useful even if the link carries a nofollow attribute, because nofollow changes how the link is counted for ranking, not whether the page’s text can be retrieved and read.
Bought links put the first step at risk
Because links still feed ranking, the old temptations still apply, and both search engines tie them to their AI features. Google’s spam policies define link spam as links created primarily to manipulate rankings, including paying for links or for posts that contain links, and sending someone a product in exchange for a write-up with a link. Bing’s guidelines list link buying, link spamming, private networks and artificial social promotion among the practices that reduce ranking eligibility and grounding visibility across Bing and Copilot.
A placement that gets a site demoted does more than waste the fee. It takes pages out of the pool the answer is drawn from, which is the only place a link was ever going to help.
Links do not explain every answer
Be careful about what links cannot tell you. When an assistant answers from what the model already knows, without searching, nothing you build today changes a model that has already been released. Links also do not explain differences between engines, because each one searches its own mix of indexes. And any study that compares link counts with AI visibility runs into the same confound: brands with many links usually have many mentions too, so a correlation cannot tell you which of the two moved the answer.
The more practical starting point is the set of pages that answers already cite for your buying questions. In Obility’s Answer Engine Insights, each observation keeps the answer together with the competitors it recommends and the sources it cites, so you can see which third-party pages the engines read in your category and whether those pages say anything about you. A frequently cited page that describes your competitors and not you is a better use of your time than a link with a higher authority score.