SEO · AI SEARCH · 8 MIN

Link Building Won't Get You Into AI Answers. Community Signals Will.

AUG 25, 2026

Filed under Seo

Two articles landed on Search Engine Land this month, one day apart. Separately they read like tactics posts. Together they redraw where AI visibility actually comes from, and the picture is uncomfortable for an industry that still bills by the backlink.

On August 12, Kevin Indig published data showing that community platforms are the largest third-party source inside SaaS-related AI answers, holding more than four times the citation share of publishers. On August 13, Evelina Milenova, who has spent six years acquiring links for 90+ SaaS brands, explained why link building only pays off once you stop treating the link itself as the product. The SEO industry is still selling one of those things. The data says the money is in the other.

What the two stories actually say

The community signals number comes from an analysis Indig ran for G2: roughly 35,000 ChatGPT citation URLs captured in Profound, US only, December 2025, with SaaS-related prompts sorted into four journey stages. Strip out the vendor's own domains and look at what remains:

• UGC platforms (Reddit, Wikipedia, Quora, YouTube, LinkedIn) hold 17.1% of cited domains overall
• Publishers hold 4.0%
• Review platforms swing with the journey: 7.4% at discovery, 13.2% at evaluation, back down to 8.4% when a buyer is comparing two finalists

UGC barely moves. It sits at 17.8%, 18.2%, 15.1%, and 17.2% across those same four stages. Review sites are a bottom-of-funnel lever you can time against a quarter. Community presence is a floor that holds everywhere, which is exactly why it cannot be bought as a campaign.

Milenova's piece makes the complementary point from the link side. Backlinks are still a confirmed Google ranking factor, and she is not arguing otherwise. But she has watched ChatGPT pull information from sites with no domain authority and no organic traffic, and LLMs plainly do not weigh links the way Google does. She cites an Ahrefs study finding that AI Overview visibility correlates with brand mentions at 0.664, against 0.218 for backlinks. The mention does more AI work than the link. Her fix is a blended model: link building for commercial pages, self-earning linkable assets, and digital PR for the story-driven coverage that produces mentions whether or not it produces links.

The budget is pointed at the wrong bucket

Here is the part that should sting. Digital PR targets publishers. Review campaigns target review platforms. Those are the two smaller buckets in the citation data, and they are where most authority-building budgets go, because marketers know what a plan for them looks like. The largest outside source class in the sample gets the least deliberate investment. Indig's own phrase: most marketing pros are not sure what a plan for it even looks like.

Consider what the prompts tell you about how buyers actually ask. More than half of the SaaS-related prompts in the study used commercial language, but only 1.5% named a vendor brand. Buyers are asking "what should I use for X" long before they ask "is Vendor Y any good." If your off-page strategy only activates when someone searches your name, you arrive after the answer has been assembled.

And the composition of that 17% floor is not what most people assume. Wikipedia, Reddit, and LinkedIn account for 99% of UGC citations in the sample. Wikipedia alone carries 10 to 14 points of the 17-point floor, and it is the one place where deliberate action is least available to you. You do not campaign on Wikipedia. You make sure the trade press, original research, and primary documentation its editors rely on exist and are accurate.

Then Reddit blinked

One week after the community signals story, the floor moved. Promptwatch reported that Reddit's share of ChatGPT Search citations fell 86.4% in four days: an average of 3.83% between July 18 and August 7, below 1% on August 14, and 0.52% through August 17. The first dip started August 8, the same day ChatGPT Search reportedly changed its query fan-out behavior. Google's products showed no comparable cliff; Reddit's share of AI Overview citations drifted from about 2.5% to 2.1% over the same stretch.

Promptwatch is careful to say the data shows when, not why, and that a collection issue is not ruled out. Treat the exact number as provisional. The pattern is still instructive, because this is not the first wobble. Indig's Growth Memo tracking had Reddit down 11.7% visibility and 10.9% AI mentions in the 28 days to June 8, then back up 18.1% three weeks later while LinkedIn jumped 43.3%. In the same windows, those platforms lost AI Overview citations while gaining organic visibility. One engine rewarding you does not mean the other one will.

That volatility only kills the lazy version of the thesis, the one where an agency sells you "Reddit SEO" as the new link building and calls it a day. The floor exists at the bucket level, not the platform level. A brand that bet its whole community effort on one platform in January would have spent August explaining a chart. The aggregate share is the number to defend. The composition is what you rebalance.

What to actually do

Indig's piece and the volatility data together produce a fairly specific playbook:

1. Find which platforms your category feeds. Run your highest-intent prompts in the engines your buyers use and record which UGC domains appear. In some categories it is Reddit and YouTube. In others it is a Discord server, a Stack Exchange site, or one trade forum. The 17% average is a bucket; the composition is local to your topic. 2. Show up as a person, not a logo. A named author with a byline outperforms the same content under a brand account in Indig's third-party signals research, and LinkedIn's own testing pointed the same way. Communities treat a person as a participant and a logo as an advertiser. 3. Answer the questions your support queue already sees. Ticket logs, sales call transcripts, and in-app search hold the exact phrasing buyers use. In a forum, that question is already asked and sitting there unanswered. 4. Correct the record in place. A 2023 thread claiming you lack a feature you shipped last year stays retrievable and keeps feeding answers. Reply in the thread with a dated correction and a link to the change; removal requests rarely land and deleting the thread deletes the context AI reads from. Set alerts on brand plus category terms. F5Bot covers Reddit and Hacker News, Syften covers Discord and Slack. 5. Pull customer language onto your own pages. One of Indig's related studies found 44.2% of citations come from the first 30% of a page. Real third-party review quotes, with links to source, belong near the top of your most-cited pages. 6. Keep link building, and change what you ask for. Milenova's blended model still applies: explicit asks for links to commercial pages, assets that earn links on their own, and PR pitched at the story rather than the link. Given the correlation data, requesting the brand mention even when no link follows is no longer a consolation prize.

What you do not do is buy reviews, astroturf threads, or rent an army of accounts. Indig says it plainly: short-term tactic, long-term brand damage. The strongest community signal is the one you did not write, which means giving customers and power users something worth posting on their own. Early access to data, a free tool, a number they can quote in an argument. That is as close to ownership as this bucket gets.

Verdict

More from the directory: Madgicx