AI quotes journalists, not your blog. Here is how to get in front of one for free.

Here is a number that should ruin your morning.

Muck Rack's Generative Pulse team pulled apart more than 25 million links that ChatGPT, Claude and Gemini handed out across 17 industries. Earned media accounted for 84% of every citation. Paid and advertorial content accounted for 0.3%. Journalism on its own made up 27%.

Read that again. Nought point three percent. The thing most small businesses spend money on, the sponsored post, the paid placement, the advertorial dressed up as a feature, is statistical noise to the machine that now decides whether anyone hears your name.

And this is not a one-off wobble. Across three editions of the study going back to July 2025, earned media has sat between 82% and 89%. Journalism has stayed between 25% and 27%. That is not a model update. That is a design decision.

So while you were writing blog post number 47, the actual game moved. The machine is not reading your site to decide if you are any good. It is reading what other people wrote about you. If nobody wrote anything, you do not exist.

What digital PR actually is, stripped of the jargon

Digital PR is getting other people to write about you, on purpose, at scale, on sites the machines already trust. That is it. No press agency lunch. No glossy campaign. No six month retainer.

It is not a press release fired into a distribution wire. It is not paying a blog $200 to slap your link in paragraph nine. It is not a guest post swap with three mates in your niche.

Here is the difference in one line: a guest post is you writing about you. Digital PR is a journalist writing about you. The machine can tell the difference, and the data above proves it weights them nowhere near the same.

Most people running a small business hear "PR" and picture a budget they do not have. Fair. But the cheapest version of this costs zero dollars and about twenty minutes a day, and I will show you exactly where it lives further down.

Why the machine trusts a journalist more than it trusts you

Two things make a source credible to an AI model, and neither of them is how good your copywriter is.

Independence. A claim about your business made by somebody with no stake in your success is worth more than the identical claim on your own homepage. Your "award winning" banner is marketing. A trade magazine calling you award winning is evidence.

Authority. Sources that get fact checked by organisations with reputations to protect carry more weight than sources that do not. Boring, unfair, and completely true.

Semrush laid the mechanism out plainly in July 2026. Patagonia has over 87,000 AI mentions in the United States. The reason is not Patagonia's website. It is that Reddit, OutdoorGearLab, REI, SwitchbackTravel, GearJunkie, TreelineReview and BetterTrail mention Patagonia more than 103,000 times between them. Shopify has over 45,000 AI mentions off the back of 44,500+ Reddit mentions, 15,400+ on LinkedIn, 1,100+ on G2 and 800+ on Capterra.

Neither brand won that by publishing more pages. They won it by existing loudly everywhere else.

This is the second of our ten AI search truths and the one people fight hardest: the machine trusts everyone except you. Your own site is the least persuasive thing you own. We drill this properly in the Dojo because it is the single belief that changes what people do on Monday morning.

The free door: HARO came back, and nobody told you

Quick history, because most SEO blogs still have this wrong.

HARO, Help A Reporter Out, was the free service where journalists posted what they needed and experts pitched answers. Cision rebranded it to Connectively in 2024 and bolted paid tiers onto it, roughly $29 to $149 a month, plus pay per pitch. Users walked. Connectively shut down on 9 December 2024.

Then Featured.com bought the HARO brand from Cision and relaunched it free on 22 April 2025, back to the original three emails a day format with AI and spam filtering added.

So the most direct route to being quoted by a journalist is free, has been free since April 2025, and a huge chunk of the market still thinks it is dead. That gap is your window. It will not stay open forever, because gaps like this never do.

How to actually win a query instead of shouting into a void

Free does not mean easy. MentionAgent's breakdown of the reactive PR workflow puts the honest numbers at 5 to 15 minutes to scan each digest, 10 to 30 minutes to write a decent pitch, and 10 to 20 pitches before one gets used. Popular queries pull hundreds of responses inside the first hour.

Which means the people who win are not the best writers. They are the fastest, narrowest and most specific. Four rules:

1. Answer within the hour. A journalist on deadline uses the first usable quote, not the best one that turns up at 6pm.
2. Pitch only what you genuinely do. A damp proofing specialist answering a query about damp proofing wins. The same person answering a query about "home improvement trends" loses to an actual analyst. Narrow beats broad every time.
3. Lead with the quote, not the intro. Give them two or three sentences they can paste straight in, in quote marks, then your name, job title and one line of credentials underneath. No "I hope this email finds you well". They are not reading it.
4. Bring a number or a specific. "We see this in about one in five properties we survey in the North East" is quotable. "It is very common" is not.

Semrush makes the same point from the other side: generic commentary is less useful, and less quotable, than specific evidence backed insight. The journalist is not doing you a favour. You are doing them one. Write like it.

The five plays that build your third party record

Reactive pitching is play one of five. Run all five and you build what the models actually read: a record of your business written by other people.

1. Publish something only you could publish

Journalists do not want your opinion. They want a number nobody else has. You are sitting on data: job counts, quote values, failure rates, seasonal patterns, the thing you see every week that the public does not.

Semrush tracked a Houzz renovation trends study that picked up 183 backlinks from 81 referring domains and got cited by AI platforms across more than 180 prompts. One page. One dataset. That is the whole mechanism.

You do not need a national survey. You need one honest number from your own books, written up clearly, on a page that says what it is. Aim for one of these per quarter and you outpace every competitor still publishing "5 tips" posts.

2. Get into other people's lists

AI systems quote "best X in Y" roundups constantly, and they quote the third party version, not yours. Semrush caught ChatGPT citing one third party roundup six times in a single answer about accounting software, without citing any of the recommended brands' own sites once.

So stop writing your own "best" page with yourself at number one. Lily Ray at Amsive has shown Google's AI Overviews will happily cite your comparison page and then recommend a competitor named inside it. You paid for the page. They got the sale.

Find the roundups that already rank and already get cited for your category and city, then email the author with a reason you belong on it. We cover the exact outreach template for this inside the Dojo, and there is a fuller breakdown in our guide on getting into AI listicles.

3. Fix your review profiles, because this one is absurd

Trustpilot and Seer Interactive studied over 800,000 AI responses across ChatGPT, Gemini, Perplexity and Google AI Mode, using more than 15,000 prompts that mirror a real buying journey.

Brands with no Trustpilot profile were cited in 1% of responses. Brands with a claimed profile and a median of 13 reviews were cited in 54%. Brands with active profiles and a median of 81 reviews were cited in 75%.

Claiming a profile and collecting thirteen reviews is an afternoon of work and a follow up email template. This is the bit agencies quietly charge $2,000 a month for and call "reputation management". It is a form and some emails.

Kevin Indig's analysis published by G2 points the same way for software: categories with 10% more G2 reviews tend to see around 2% more AI citations. Different platform, same lesson. Reviews are not vanity. They are the corroboration layer.

4. Say the same thing about yourself everywhere

Wikipedia, Wikidata, Crunchbase, Google Business Profile, LinkedIn, YouTube, G2, Yelp. Pick the ones that apply to you and make the core facts identical across all of them: name, category, founding date, what you actually sell, where you operate.

You are an entity, not a keyword. If three sources call you a "marketing agency" and two call you an "SEO consultancy", the machine has two half confident guesses instead of one certainty. Boring work. Massive payoff. Nobody does it.

5. Match the outlet to the platform your buyers use

The three big models are not one audience. From the same Muck Rack study: ChatGPT cites sources in 96% of responses but averages only five citations. Gemini cites in 82% of responses, averaging eight. Claude is the pickiest, citing in 55% of responses but averaging thirteen sources when it does.

Their favourite domains differ too. ChatGPT leans on Wikipedia. Claude leans on PubMed Central. Gemini leans on Reddit. Journalism citations spread across more than 20,000 distinct outlets, though Axios turns up in ChatGPT's top three cited domains in 13 of 17 industries, the only journalism outlet to manage that anywhere.

Practical version: if your buyers live in ChatGPT, encyclopedic and reference sources matter more. If they live in Gemini, Reddit matters more. A pitch strategy tuned for one will not land the same on another.

What this actually costs you

Cash is the small number here. Time is the real invoice.

I am not going to pretend I have live pricing for every platform in this space, because it moves constantly and half the "2026 pricing" pages out there are guessing. Check the source before you pay. That goes for us too.

Your first 30 days, in order

Days 1 to 3. Claim your review profiles. Trustpilot if you sell to consumers, G2 or Capterra if you sell software, Google Business Profile always. Get the details identical across every one. Then build a one line request email and send it to your last twenty happy customers. You are chasing thirteen reviews, not a hundred.

Days 4 to 7. Sign up to HARO at helpareporter.com. Free, email only. Spend the week just reading the digests without pitching, so you learn what a real query looks like and how fast they move.

Days 8 to 14. Start pitching. Two queries a day, both dead in your lane, both answered inside the hour. Track every pitch in a sheet: date, outlet, query, sent time, outcome. Most people skip the tracking and then cannot tell you what works.

Days 15 to 21. Entity clean up. Wikidata, Crunchbase, LinkedIn, YouTube description, your Google Business Profile category. Same name, same category, same founding date, same description everywhere.

Days 22 to 30. Pick your data piece. One number from your own business, written up on one page with a clear title and a clean methodology paragraph. Ship it. Then send it to five journalists who have covered that beat this year.

That is thirty days and zero dollars of platform spend. It is also about twelve hours of work, which is why most of your competitors will read this and do nothing.

Who this suits, and who it does not

This suits you if you actually know something. Trades with real field experience, consultants, clinicians, specialists, anyone who sees a pattern the public does not. Journalists need exactly that and cannot get it from a chatbot.

This does not suit you if you are a pure reseller with no distinctive view, if you cannot commit two hours a week for three months, or if you need leads by Friday. Digital PR compounds. It does not sprint. If you need volume this quarter, run paid alongside it and read our SEO versus PPC breakdown before you decide the split.

The mistakes that burn six months

• Pitching everything. Spray and pray gets you blacklisted by the journalists you most want. Narrow wins.
• Chasing DR instead of relevance. A trade title your buyers read beats a general site with a bigger number next to it. The models cite across 20,000+ outlets, not five.
• Buying the placement. Paid and advertorial content is 0.3% of AI citations. You are buying the least valuable version of the thing.
• Writing your own "best of" page and calling it PR. You may end up promoting a competitor inside your own content.
• Measuring placements instead of citations. A clipping book is not a result. Track citation lift and share of voice in AI answers, and track referral traffic properly. Our guide on tracking AI traffic in GA4 covers the plumbing.
• Stopping after one hit. One quote in one outlet does not move an entity. Twelve do.

Alternatives, if pitching makes you want to lie down

Not everyone will do reactive PR. Fine. Three other routes to the same destination:

Reviews only. The Trustpilot and Seer numbers say a claimed, active profile does a serious amount of the work by itself. Lowest effort, highest floor.

Community presence. Gemini's most cited domain is Reddit. Being genuinely useful in the right subreddit, over months, with no link dropping, builds a third party record the models read. Spam it and you get banned and deserve it.

Classic link building. Still works, still matters for Google, just stop expecting it to carry AI visibility on its own. We broke the split down in backlinks versus brand mentions, and if you want the real market rates before anyone quotes you, read how much backlinks actually cost. If you would rather we just ran it, that is what our link building service is for.

The risks nobody selling you PR will mention

Coverage is not control. A journalist can quote you accurately and still frame the piece in a way you hate. If your opinion is spicy, expect that occasionally.

Links are not guaranteed. Plenty of outlets will name you without linking. That still builds the entity record, but if you are measuring in links alone you will call a win a loss.

Timelines are genuinely unpredictable. Semrush's own answer is days to months, depending on whether the platform searches the live web or waits for a model update. Anyone promising you AI citations by a specific date is making it up.

And the biggest one: this works because most businesses are not doing it. That advantage decays. The people who start in the next six months get the cheap version. Everyone after that pays more for less.

Where this goes next

Three editions of the Muck Rack study, stretching from July 2025 to May 2026, and earned media has not dropped below 82%. That stability is the signal. These systems are not going to suddenly start trusting your homepage. If anything, as AI generated content floods the open web, independently verified human sources get more valuable, not less.

Expect two things. Journalist request platforms will get more crowded and more filtered, so quality of pitch beats volume harder every quarter. And review and reference platforms will get weighted even more heavily, because they are the cheapest corroboration signal a model can check.

The move does not change. Get other people to say true things about you, in places the machine already reads. Start with the free door.

Right, go and do it

Claim your review profiles today. Sign up to HARO tonight. Pitch two queries tomorrow. That is the whole first week and it costs nothing but your attention.

If you want the pitch templates, the outreach scripts, the tracker and someone checking your work instead of you guessing for six months, that is what the Dojo is. Every play in this article, broken into steps, with the templates done.

Join The Dojo. $67/mo. Or if you would rather we ran the whole thing while you get on with the actual business, take a look at Done For You.