Ask three SEO professionals whether backlinks still matter for AI search and you will probably get three confident answers. One will say links are dead and mentions are everything. Another will say nothing has changed. A third will point to a study that supports whichever view they already hold.
The data is more interesting than any of those positions. The 2026 studies do not agree, and the reasons they disagree tell you something useful. Backlinks still matter for Google rankings. How much they matter for AI answers depends on which AI system you mean, which query you test, and whether you count a link or a mention. A brand that treats all of that as one question will spend money in the wrong places.
This article sets out what the evidence shows, where it conflicts, and what a business should do with a limited budget.
What the studies say, and where they disagree
Five research sources dominate the current conversation. They measure different things, so they are not directly comparable, but laid side by side they show the pattern.
| Study | What it measured | Main finding on backlinks | Main finding on mentions or earned media |
|---|---|---|---|
| Ahrefs, 75,000 brands (December 2025 data) | Correlation with brand visibility in Google AI Overviews | Backlink count correlated weakly, at about 0.22 in the widely quoted figures | Branded web mentions correlated much more strongly, at about 0.66. YouTube mentions were higher still in the same analysis |
| Muck Rack, “What Is AI Reading?” (May 2026 edition) | Sources cited across more than 25 million links in ChatGPT, Claude and Gemini responses | Not measured as a ranking signal | About 84% of citations were earned sources, as defined by the study’s broad category. Paid and advertorial content was a small fraction |
| Salt.agency (undated in the version reviewed) | Correlation of domain metrics with visibility in four AI tools | Moderate correlation for ChatGPT, Perplexity and Gemini. Weak for Google AI Overviews | Not the study’s focus |
| Semrush with Kevin Indig, 1,000 domains | How often domains were mentioned across five AI platforms, correlated with backlink metrics | Moderate to strong relationships overall. Strongest for ChatGPT Search and Perplexity. Weakest for Google AI Overviews | High-quality mentions were reported to matter even when they are nofollow |
| HubSpot analysis (reported by Mediaweek, July 2026) | Thousands of high-ranking Google pages checked against AI visibility | No relationship found between backlink strength and AI visibility | Pages with limited authority still appeared when they answered a question clearly |
Read side by side, the picture is clear enough to act on, and messy enough to deserve caution. Three points stand out.
First, the strongest signals in the Ahrefs study are about mentions, not links. Brand names appearing across the web, and YouTube mentions in particular, correlated more strongly with Google AI Overview visibility than backlink counts did. Correlation is not causation, and the figures vary between versions of the analysis, so these numbers are best read as a direction, not a precise measure.
Second, the platform matters. Several studies found that Google AI Overviews respond to links less than ChatGPT and Perplexity do. If your audience uses ChatGPT for research, a link-based view of authority may still be relevant. If your traffic is mostly from Google AI Overviews, the link signal may be weaker.
Third, the HubSpot finding is the outlier that should worry anyone who has spent years building links. Its public methodology is not fully available, and it is reported through an interview summary rather than a published dataset. It should not be treated as proof that links have no role. It is a reminder that the relationship can be close to zero in some samples.
Why the studies disagree
The disagreement is not a sign that someone is lying. Several structural differences explain most of it.
They measure different outcomes. Some studies measure whether a brand appears in an AI Overview. Others measure whether a page is cited, whether a brand is named, or whether a domain appears across many prompts. A brand can be named without a link, and a page can be linked without being cited.
They use different prompts. A study built on broad informational questions will show different results from one built on commercial comparisons such as “best payroll software for a 50-person company”. Muck Rack’s analysis found that citation patterns vary by query type, and that some verticals behave very differently from others.
The top-10 overlap is shrinking. In mid-2025, Ahrefs found that about 76% of pages cited in Google AI Overviews also ranked in the top 10 for the same query. Reporting on later data suggests that overlap has fallen sharply, with one summary citing about 38% by early 2026 and another analysis putting it lower still. Because the method varies, the exact figures are contested. The direction, though, is consistent: AI systems are drawing from a wider pool than the traditional top 10, and a link profile that works for rankings does not automatically carry over.
The sources are not neutral. Several studies come from companies that sell link building, digital PR, or AI visibility tools. Some have a commercial reason to show that links matter, and others have a reason to show that they do not. This does not make their data wrong. It does mean the methods, sample sizes and definitions deserve a close look before anyone builds a budget around a headline number.
What the evidence agrees on
Despite the disagreement, four findings appear again and again.
1. AI answers lean heavily on third-party sources. Muck Rack’s May 2026 analysis found that earned sources made up about 84% of citations across ChatGPT, Claude and Gemini, under its broad definition of earned media. The company’s October 2026 edition, drawing on more than 40 million AI responses, kept earned and third-party sources at roughly the same share, while journalism’s share dipped from about 27% to 22%. The pattern is durable across editions. Your own website is rarely the main source of the citation.
2. Being named matters as much as being linked. Across the studies, unlinked brand mentions repeatedly appear as strong signals. An AI system can name your company, recommend your product, and cite a third-party page about you without ever linking to your domain.
3. Some platforms are shaped by community sources. Gemini’s most-cited domain in the May 2026 Muck Rack edition was Reddit, at about 2.4% of all citations. Yet the share can move sharply. One analysis reported that Reddit’s share of ChatGPT Search citations fell by 86% over four days in August 2026. Community sources can matter a great deal, and they can also disappear from view without warning.
4. Results change quickly. AI answers vary between runs, between days, and between accounts. A single test is an anecdote. A repeated test across a fixed set of prompts is a measurement.
The three jobs a backlink still does
It is more useful to ask what a backlink does than whether it “matters”. In 2026, a link does three jobs, and only one of them is about AI.
Job one: it helps a page rank
Google’s ranking systems still use links as one signal of authority and relevance. Google’s own guidance on AI features states that established SEO practice remains the foundation. If a page ranks better because of its link profile, it has a better chance of being crawled, indexed and considered by systems that pull from the top results. This indirect path is real, and it is the reason links should not be abandoned.
Job two: it sends referral traffic
A link from a trusted site can send visits directly. This matters more for some sites than others, and it is measurable in analytics. It is also the easiest of the three jobs to evaluate.
Job three: it signals that a third party vouches for you
This is the job most closely tied to AI visibility, and it is the job that a link shares with a mention. When a respected publication, academic source or industry body refers to your company, that reference can help an AI system decide whether your brand is a credible entity to name. The link is one form of that reference. An unlinked mention in the right place can do part of the same work.
The practical conclusion is that links are not being replaced by mentions. They are being joined by them. A serious off-site programme now needs to earn both.
What digital PR looks like when AI is the audience
Digital PR, meaning earned coverage in news outlets, trade publications and industry sources, has become the discipline most often recommended for both links and AI visibility. Most of the advice in circulation is generic, so it helps to be specific about what works.
Publish original data that others want to cite. A survey, a benchmark or an analysis of a public dataset gives journalists and researchers something specific to reference. The data must be accurate, the method must be stated, and the limitations must be clear. Invented numbers are a liability, and AI systems and editors both penalise weak sourcing.
Give journalists a quotable expert. Commentary from a named specialist, with a clear point of view, is easier to cite than an anonymous company statement. This is the kind of expertise that helps an AI system judge whether a source knows what it is talking about.
Target lists and comparison pages. In a March 2026 study of 75,000 AI answers reported by Search Engine Land, listicles accounted for a large share of citations on commercial queries. If your category has well-read comparison pages, getting included in the right ones may be worth more than a dozen generic guest posts.
Be careful with press releases. Muck Rack’s data, as summarised in several 2026 reports, puts press releases at about 1% of AI citations, despite their wide use. Distributing a release is rarely a strategy by itself. It needs a story that a journalist would pursue without it.
Treat community platforms with care. Reddit and YouTube can help, but they reward genuine participation. Promotional posts are often removed, and they damage the trust they are supposed to build.
How to measure links and mentions separately
Most teams measure AI visibility badly because they measure one number. A better approach separates the outcomes.
Step 1: Build a fixed prompt set. Choose 30 to 50 questions your buyers actually ask. Include informational prompts (“how do I choose a payroll system”) and commercial prompts (“best payroll software for a 50-person company”). Keep the list stable for at least three months.
Step 2: Test across the platforms your buyers use. Run the same prompts in ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. Record whether your brand is named, whether your domain is cited, which third-party sources appear, and whether the answer recommends you.
Step 3: Repeat under comparable conditions. Run the test weekly or fortnightly, and note the date, the account type and any settings. Averages over time matter more than any single result.
Step 4: Track the earned sources. For each citation, record the domain. Over time, you will see which publications, communities and reference sites keep appearing. Those are your off-site targets.
Step 5: Connect it to business outcomes. Compare AI visibility with branded search volume, direct traffic, referral traffic from AI tools where it is visible, and enquiries. Google Search Console does not currently separate most AI Overview clicks from standard organic clicks, so treat any split as an estimate.
Five mistakes to avoid
Buying links in volume. Low-quality link schemes carry ranking risk and are unlikely to build the editorial credibility that AI systems appear to reward.
Treating correlation as cause. A strong correlation between mentions and AI visibility does not prove that a particular mention caused a citation. It is a reason to test, not a guarantee.
Ignoring the platform. A tactic that helps your visibility in ChatGPT may do little for Google AI Overviews, and the reverse can also be true.
Measuring only one metric. A rise in brand mentions with no change in enquiries is a signal to investigate, not a success.
Assuming a stable landscape. Reddit’s citation share can collapse within days, and Google’s updates can move rankings in a matter of weeks. Plans should be reviewed quarterly.
A practical 90-day plan
Days 1 to 30: establish the baseline. Audit your current backlink profile, record your brand mentions across the web, and build the prompt set. Check whether your core pages rank and are indexed, because nothing else works without that.
Days 31 to 60: create one piece of original evidence. A focused study, benchmark or dataset analysis in your niche gives journalists and editors a reason to write about you. Pitch it to a small list of relevant publications and specialists, not a mass distribution.
Days 61 to 90: build the presence that AI systems read. Identify the comparison pages, industry lists and community discussions that appear in your prompt results. Work on accurate, specific inclusion where it is legitimate: update your own product or service pages, supply accurate information to listings you can influence, and contribute genuinely useful expertise where your audience already discusses the topic.
At the end of 90 days, compare the results with the baseline. If mentions rose and enquiries did not, look at the landing pages and the match between the content and the questions being asked. If neither moved, the problem is probably the evidence itself, not the outreach.
How SarkarSEO approaches this
At SarkarSEO, we treat backlinks, mentions and on-site content as parts of one plan, rather than as competing strategies. Link building still has a clear role where it supports rankings, and we are careful about which links we pursue and how they are earned. We also work on the questions AI tools are asked about a client’s category, the evidence on the client’s own pages, and the consistency of how the brand is described elsewhere.
We are happy to talk through what a realistic plan would look like for your business, including what we would measure and what we would not promise. If you want to discuss off-site SEO or AI search visibility, you can see our premium SEO services or get in touch with our team. For a broader introduction to how these disciplines fit together, read our post on the difference between SEO, GEO and AEO.
Frequently asked questions
Do backlinks still help my Google rankings?
Yes. Google’s systems still use links as one signal among many. The studies that question links in AI search are generally about AI answers, not about traditional rankings. Keep building links for rankings, and judge their AI effect separately.
Are unlinked brand mentions really more valuable than links?
For AI visibility, several studies suggest they can be a stronger signal than backlink counts. For rankings, links remain important. The two do different jobs, so the right answer depends on the outcome you are measuring.
Why does Google AI Overviews behave differently from ChatGPT?
Each system draws on different data, retrieval methods and ranking inputs. Google’s AI Overviews are closely tied to Google’s own search results, while ChatGPT and Perplexity may draw on other sources, including community and reference sites. Studies consistently show that the platforms differ, so measure them separately.
Is digital PR the same as link building?
No. Digital PR earns coverage in publications, and links and mentions often follow from that coverage. Link building starts from a target link. Digital PR starts from a story. Both can earn links, but digital PR is more likely to produce the third-party mentions that AI systems read.
How long does it take to see a change in AI visibility?
There is no reliable fixed timeline. Results vary by platform, query and category, and AI answers can change from one day to the next. Measure over at least three months, using a stable prompt set, before drawing conclusions.
Tell us about your website, your category and the questions your buyers ask. We will review your current links, mentions and on-site evidence, and suggest a realistic next step. Contact SarkarSEO to start.
Sources and further reading
- Ahrefs, “Top brand visibility factors in ChatGPT, AI Mode and AI Overviews (75K brands studied).” ahrefs.com/blog/ai-brand-visibility-correlations (primary source; figures reported differently across versions).
- Muck Rack, “What Is AI Reading?” May 2026 edition. Read the report (PDF) (primary source).
- Muck Rack, “What Is AI Reading?” October 2026 edition, as reported by Unite.AI, October 28, 2026. unite.ai (secondary report).
- HubSpot analysis by Aja Frost, reported by Mediaweek, July 2, 2026. mediaweek.com.au (secondary report; no public dataset located).
- Semrush and Kevin Indig, study of 1,000 domains across five AI platforms. semrush.com/blog/backlinks-ai-search-study (primary source).
- Salt.agency, “Backlinks aren’t dead, but they’re not enough in AI search.” salt.agency (undated in the version reviewed).
- Google Search Central, “AI features and your website.” developers.google.com/search/docs/appearance/ai-features (primary source).
- SarkarSEO, “SEO vs GEO vs AEO: A Simple Guide.” Read the post (internal reference).
