Guide
Where do AI assistants get the answers they cite?
Brett Farmiloe runs HARO, Connectively and Featured, and measured 22,000 AI citations across 405 audits. 64% came from earned media rather than a brand's own website, and 34% came from sites with a domain authority under 40.
The short answer
Mostly from editorial coverage on other people's sites, not from your own. Brett Farmiloe — CEO of HARO and founder of Featured — ran an AI citation report across 22,000 citations from 405 generative-engine-optimization audits, and found 64% came from earned media. More surprising is where that coverage sits: 34% of citations pointed to websites with a domain authority under 40, the small blogs and page-two-through-ten results that no PR agency puts on a media list. His summary of the gap is that 86% of brands aren't mentioned for the prompts and queries they most want to be mentioned for. The practical consequence is that optimising your own site is the smaller half of the job, and the accessible third of the citation pool is the part almost nobody competes for.
Step by step
Run a GEO audit before you pitch anything ▶12:41
Farmiloe's first move is a generative engine optimization audit: find out which sites AI engines actually cite for your topic, then work out how to get into those specific sites. His own company offers one free, and others exist. The output he describes is a prioritised list of real pages — an example he gives is a small RV-rental blog carrying a listicle that does not mention your brand — along with the editor to contact and a reason for the outreach.
Go after the long tail, not the tier-one logos ▶20:36
Traditional PR chases Forbes, Fortune, Fast Company and Inc. Farmiloe's data says the page-two-through-ten results now matter, because a third of all citations point to domains scoring under 40 for authority. That third is, in his words, wide open: it is not what a retained PR firm is targeting, and a small site covering your exact niche is reachable by one email.
Make the pitch helpful, authentic, relevant and on time ▶11:03
He uses the HARO acronym as the test for a pitch. Helpful: are you making the journalist's job easier. Authentic: is there a human behind it. Relevant: journalists state exactly what they need, down to the required job title and the subject line, and the single most common failure is answering when you do not qualify. On time: they are on deadline and usually take the best source who arrives first.
Use AI to polish, never to generate and submit ▶14:54
Every pitch entering HARO runs through an AI detector. 21% come back 100% AI-generated, while 35% of journalists one-click opt out of anything AI-written. Farmiloe's distinction is concrete: dictating your own experience into a model and asking it to tidy the wording is AI-assisted and fine, because the insight is still yours. Having a model read the queries and auto-submit without you looking is spam. He describes a journalist's inbox as among the best-trained AI detectors there is.
Pitch creators, not only journalists ▶34:33
The ratio has inverted. Farmiloe counts a little more than six PR professionals for every working journalist, and about thirty full-time creators for every journalist — roughly 1.5 million people whose job is publishing. That leaves about five creators for every PR person, which is the opposite of the competition you face for a journalist's attention, and those creators actively need new material.
Treat it as consistency, not a campaign ▶48:39
His test for whether it is working is a customer saying they found you because an AI recommended you. The reason it has to be ongoing is that the ground moves: he points to Reddit going from a favoured source in AI answers to an 80% drop in how often it was referenced, effectively overnight. The question he frames the work around is how you survive the next model update, and how you get into all of the engines rather than one.
“34% of AI citations were from websites with domain authorities of less than 40”
Nick Standlea, in the video
Key moments
- ▶0:57They shut HARO down. He bought it.
- ▶3:1764% of AI citations come from earned media
- ▶11:03A winning pitch: helpful, authentic, relevant, on time
- ▶12:41Your first move this week: run a GEO audit
- ▶14:5421% of pitches are pure AI. 35% of journalists opt out.
- ▶20:36Why page two through ten suddenly matters
- ▶34:33Thirty full-time creators for every working journalist
- ▶48:39Surviving the next model update
Questions people ask
where do ai assistants get their information
Largely from editorial coverage published on other sites. Brett Farmiloe's AI citation report measured 22,000 citations across 405 audits and found 64% came from earned media rather than brands' own websites. He also notes a Cloudflare finding that more bots than humans now read web pages, and that publishers are increasingly optimising for machines because the destination has changed.
how do i get my business mentioned by chatgpt or claude
Get covered on the sites those engines already cite for your topic. Farmiloe's sequence is to run a generative engine optimization audit to find which sites are being cited, work out how to get into those specific sites, then pitch them. He is explicit that the first step is research rather than outreach, because the sites that matter are rarely the ones you would have guessed.
what percentage of ai citations come from small websites
34%, in Featured's audit of 22,881 citations — those pointed to domains with an authority score under 40, and 13.2% to domains under 20. Worth scoping it honestly: that figure is from Perplexity citations, drawn from 405 audits run on Featured's own platform between June and August 2026, covering commercial buying questions. It is a large real sample, not a neutral census of the whole web.
what is a geo audit
Generative engine optimization audit — a check of which websites AI engines are citing when they answer questions in your category. Farmiloe describes the output as a prioritised list of the sites to reach out to and the reason for each approach, including the editor's contact and a draft pitch. The point is to stop guessing which coverage would help.
do journalists accept ai written pitches
Many actively refuse them. 35% of journalists coming to HARO one-click opt out of anything AI-generated, while 21% of the pitches arriving are entirely machine-written. Farmiloe says journalists are among the best-trained people anywhere at spotting it, because of the sheer volume they read — and that a pitch which wastes their time creates work rather than saving it.
what is the difference between ai assisted and ai generated
Who supplies the substance. Farmiloe's example of AI-assisted is voicing your actual experience and insight into a model and asking it to polish the response — the expertise is yours, the wording got tidied. AI-generated is letting a model ingest the queries, write answers and submit them without you reading them, which he calls spam plainly.
why does page two of google matter now
Because an AI answer is assembled from many sources rather than ranked into a list. Farmiloe's framing is that Google's job was to produce a directory and let the user choose, while an engine generating an answer may draw on dozens of sources at once. Being one of several mentions across those sources is what gets a brand into the summary, and many of those sources sit well below page one.
how many brands are invisible to ai
86% aren't mentioned for the prompts and queries they want to be mentioned for, according to Featured's AI citation report. Farmiloe's blunt summary on the show is that most brands are invisible — and the counterpart is that the accessible third of the citation pool has very little competition in it.
This guide is drawn from a full conversation on the show.
Full episode: 64% of AI citations come from here (not your website), HARO CEO Brett Farmiloe · All guides · Sponsor the show