Your Blog Isn't Where AI Finds You
Twelve blog posts last quarter. A content retainer that renews every month without anyone asking what it bought. Then a customer opens ChatGPT, asks who to hire in your category, and the answer names three businesses that are not you.
The problem is not your writing. It is where the machine looks before it answers. Analysis published August 12, 2026 by Kevin Indig in Search Engine Land, built on Profound citation data he ran for G2, found that community platforms account for 17.1% of cited domains, against 4.0% for publishers. More than four times the share, and not one of those pages was written by a marketing team. The pages deciding whether AI names you are pages you cannot write, which means the work is showing up as a named human where your category already gets discussed and correcting the record where AI already reads.
Four times more of AI's sourcing sits in rooms you cannot edit
Profound measured which domains ChatGPT actually cited across vendor-seeking prompts in December 2025, US market. Indig's read of it is blunt about the concentration: Wikipedia, Reddit and LinkedIn together make up 99% of all user-generated citations in that sample, with Wikipedia alone carrying between 10.1 and 14.0 points of the 17-point community share. Three websites. That is the community layer.
Independent work points the same direction. A Search Engine Land citation study across ChatGPT, Gemini, Perplexity and AI Overviews found Reddit is the single most-cited domain, at roughly 16.7% of ChatGPT citations. Semrush's three-month most-cited-domains study and Ahrefs' July 2026 ranking of the 50 most-cited domains in ChatGPT point at the same names: Reddit, Wikipedia, YouTube, LinkedIn and Forbes. Four separate datasets, same shape.
Indig's numbers are one engine, one country, one month, and one vertical lean: B2B SaaS buyers looking for software vendors. He notes that 91% of citations appear in only one engine, so ChatGPT is not a stand-in for AI search as a whole. Read the three pages that rank for this topic and you get the domain rankings plus advice to get listed on G2, claim your Capterra profile, and seed comparison threads. If you run a physiotherapy practice, a dental clinic, an HVAC company or a law firm, none of that exists for you. There is no G2 for a clinic in Camden. None of that advice exists for a business with no software category to be listed in, so the rest of this is the translation.
Owners overfund the blog because it is the only room they control
Sitting across from a business owner, I can predict the budget split before I see it. Most of the money goes to the website, because the website obeys. You publish, it appears. You ask for a change, it changes by Thursday. Community platforms do the opposite: you cannot publish a Reddit thread about how good you are, you cannot commission a Wikipedia entry, and the one time you try, the room notices and punishes you for it.
So the budget flows to the controllable thing, and the controllable thing is not where the citations come from. Similarweb's analysis of why AI engines cite user-generated content over brand content lands on the same uncomfortable arithmetic: your own pages read as the seller's account, and the model treats them accordingly. This is a distribution problem wearing a content costume, and it is the same failure I wrote about in why brand mentions became the currency backlinks used to be. Publishing harder does not fix a sourcing problem.
| Where the answer gets sourced | Who writes it | What you can actually do |
|---|---|---|
| Community platforms (17.1% of cited domains) | Customers, peers, strangers, volunteer editors | Participate as a named person, answer questions honestly, give customers a reason to say your name |
| Publishers (4.0% of cited domains) | Journalists and editors | Be quotable and reachable, respond to journalist requests in your category |
| Your own site | You | Everything, which is exactly why it gets weighted as the seller's own account |
There is a second reason owners get this wrong, and it is more forgivable. The blog does produce something visible. Sessions, time on page, a chart that goes up. Community presence produces nothing you can screenshot for six months, then produces demand you struggle to attribute. Given a choice between a measurable small lever and an unmeasurable large one, most people pick the one with the dashboard.
The community floor holds no matter how close your buyer is to paying
What makes this data worth changing your plan over is not the headline percentage. It is the stability. Indig split citations by buying stage and found community share barely moves: 17.8% at discovery, 18.2% at exploration, 15.1% at evaluation, 17.2% at focused evaluation. A range of 3.1 points across the whole purchase process. Review sites behave completely differently, swinging from 7.4% up to 13.2% and back down to 8.4% as intent rises and falls.
| Buying stage | Share of citations from community platforms | What it means for you |
|---|---|---|
| Discovery | 17.8% | You are named before anyone knows your category well |
| Exploration | 18.2% | Peak influence, and the stage most owners ignore entirely |
| Evaluation | 15.1% | The floor dips slightly, then recovers |
| Focused evaluation | 17.2% | Still there when the buyer is picking, not just browsing |
Indig draws the line in one sentence.
Practically, that means a community mention is not a top-of-funnel nicety you get to postpone. It is working on the buyer who is thirty seconds from calling someone, and it is working on the buyer who does not yet know your category exists. One asset, both jobs, and you are not paying per click for either.
The cost of ignoring it compounds on a slow clock, which Indig also spells out.
"Third-party authority takes a quarter to move. Proprietary data takes a research cycle. Community presence takes a year, and once it takes off organically, then holds a floor that doesn't care which stage of the journey your buyer is in."
Kevin Indig, “Community signals are AI's largest third-party source,” Search Engine Land, August 12, 2026.
A year. That is the honest price, and it is why I would rather you start badly this month than start perfectly next January.
A clinic's version of G2 is a register, a subreddit and a named person
You will not get a Wikipedia page; local businesses almost never meet the notability bar, so treat that 10 to 14 points of the community floor as terrain you do not compete on. What is left is Reddit, LinkedIn, YouTube, and the smaller rooms specific to your trade: a city subreddit, a condition-specific support forum, a trade association directory, a Facebook group for local parents, a professional register that AI systems read as a factual reference. That last category matters more for a clinic than any of the big three, and nobody writes about it.
I booked a London ADHD clinic solid for three straight months. Demand got heavy enough that they hired more specialists and outsourced the overflow. What I took from that engagement, across the medical, local services and education clients in the 300-plus businesses I have worked with since 2008, is that in categories where the decision is emotional and expensive, the shortlist forms in conversation long before anyone types a query. People ask other people first. AI systems now read those same conversations at scale, which means the room where your shortlist gets built is also the room your future customers' assistant is quoting.
Steps three and four are smaller jobs than they sound. If you run a physio practice, you can check your entry on your professional register and your association directory this afternoon: right name, right address, no practitioner who left two years ago still listed, and services described the way a patient would say them rather than the way a clinician writes them. Where something is wrong, email the registrar with the corrected details and keep the reply. For step four, the sentence you hand a happy patient is not "leave us a review" but "if you ever see someone asking about back pain in that parents' group, mention what the assessment caught."
Step two decides whether the rest works: the posting has to come from a named person, not the business account.
Indig's shortest version of the rule is that “the strongest community signal isn't written by you.” So if you delegate this to an agency, the question to ask is not "can you do Reddit." Ask them: which named human on our team will be posting, what will they say in month one before any link appears, and what is the plan when a thread turns hostile. An agency that answers with a posting schedule and a content calendar has misunderstood the assignment. This is also different from getting your own pages to outrank forums, which I covered separately in why you should stop trying to outrank Reddit in classic Google search. Different problem, different fix.
One more thing to steal from the same dataset, because it applies to every page you do control. Indig's cross-reference found 44.2% of citations come from the first 30% of a page.
Answer first, credentials second, story third. That applies to your service pages, and it applies to how you write a forum reply.
Count the rooms you appear in, not the sessions you already had
Measurement is where this work dies, because the first metric people reach for is traffic and community presence does not produce much of it. Somebody asks ChatGPT for a recommendation, gets your name from a thread, then searches your brand directly. Your analytics records a direct visit. The room that caused it is invisible.
So change what you count. Once a month, ask the assistants the questions your buyers ask, in the wording your buyers use, and record whether you are named and what source the answer leans on. Track the number of distinct third-party pages where your business appears by name. Track branded search volume, because a rising line there with flat non-branded traffic usually means something off-site is doing the introducing. Watch for the phone call that starts with "I saw you mentioned somewhere."
Set the review cadence at ninety days and hold it. If you want the wider sequence, including the on-site pieces this post deliberately leaves alone, the step-by-step AI visibility checklist covers the full order of operations.
None of this means firing your content team. It means demoting the blog from the main event to what it actually is: the page that closes the loop after something else made the introduction.
Frequently Asked Questions
Should I stop writing blog posts if AI mostly cites other sites?
No, but stop treating volume as the strategy. Your own pages still do real work: they confirm what a third party said, they answer the specific question a buyer has after the introduction, and they carry your prices, hours and services in a form machines can read. What they will not do is get you named in the first place. Keep a small number of genuinely useful pages, put the answer in the first third of each one, and move the rest of that budget toward being present where your category gets discussed.
How do I find where people actually talk about my business type?
Start with two searches on Reddit: your service plus your city, and the exact question a worried customer would ask a friend. Then check whether your trade has a professional register, an association directory or a condition-specific forum, because for clinics and licensed services those reference pages carry more weight than any social platform. Ask your last ten customers where they looked before they called you and write down the answers. That list of rooms is more useful than any tool, and it costs you an afternoon.
Can I pay someone to post about my business in these communities?
You can, and it backfires in a way that is expensive to undo. Communities are unusually good at spotting paid promotion, moderators remove it, and the removal produces a public thread about your business being caught, which then becomes a citable page about you. The version that works is slower and cheaper: a named person from your team answering questions honestly, including the questions where the right answer is that you are not the best fit. Kevin Indig's finding on this is direct, that the strongest community signal is not written by you.
The instinct I keep seeing is the same: when visibility drops, publish more. It worked for a long time, and it is now the most expensive habit in marketing. Spend an hour this week finding the ten threads where your category actually gets debated, and read what people say about businesses like yours when no salesperson is in the room. If what you find is a shortlist you are not on, or a version of your business that is two years out of date, book a call and we will work out which rooms are worth your time and which ones are noise. The question worth sitting with is simpler than any of the data above: when was the last time your business was recommended by someone who does not work for you?
