You have a few mentions out there, but the AIs ignore you when they answer the questions your customers actually ask. It's not a product problem: it's that the citations the models use don't happen by chance, and the scattered ones you have now aren't enough to build a recognizable pattern. Meanwhile, the competitors who own the right sources show up in your place — every month. There is a precise, repeatable and measurable process to earn mentions that AI recognizes as relevant: and within a year it completely changes the situation.
Open your inbox and search for “journalist pitch”. Zero results? That’s the first reason AI doesn’t cite you.
If the search comes back empty — or returns three emails from two years ago — you’re working on AI visibility with one arm tied behind your back. Text mentions on industry publications, authoritative blogs and vertical newsletters don’t arrive because your product is good: they arrive because someone asks for them, in a structured, repeatable, measurable way.
In this article I explain the monthly workflow I use with clients to turn outreach from a sporadic activity into an industrial process. The end goal is just one: to show up in the answers of ChatGPT, Perplexity, Gemini and Claude when a potential customer asks “best [industry] in [area]”.
Why AI rewards those who have an outreach process, not those who improvise
In the field of research on citations generated by language models, the work of Huang et al. (2024) “Training LMs to Generate Text with Citations via Fine-grained Rewards” documents a precise technical mechanism: LLMs are trained to generate answers by citing passages retrieved from a corpus of millions of documents, and the system rewards the citation when it is relevant and supportive with respect to the user’s question.
Translated into entrepreneur language: the model doesn’t cite a source “because it’s famous”. It cites it because, at retrieval time, it finds a text passage that matches the question and that contains a clear reference to the brand. If your name doesn’t appear inside text passages relevant to the question, the system has no material to draw from.
From this follows a brutal operational consequence for someone selling artisanal egg pasta in Teramo: the model doesn’t cite you because you exist. It cites you if there are text mentions of you, scattered across authoritative sources, that contain the keywords the customer uses to ask the question (e.g. “Abruzzo PGI pasta”, “artisanal pasta factory Teramo”, “in-house egg pasta production”).
These mentions don’t create themselves. You need a process, a pitch, a rhythm.
The monthly workflow: 20 targets, 4 touches, one metric
The process that works is simple in structure and hard in execution. I’ll sum it up in five phases.
Target identification. Every month pick 20 people — not 20 publications, 20 people. Food & wine journalists, regional cooking bloggers, travel guide editors, trade magazine writers (Gambero Rosso, Dissapore, local Abruzzo food blogs), authors of vertical newsletters. You find them by searching “Abruzzo pasta” + “journalist” on Google News over the last 12 months, or by using Google Trends to understand which topics are rising.
Personalized pitch. The pitch must contain one of these three things: a data point you’ve never told publicly, an expert comment on an industry news item published in the last 48 hours, an offer of access (a visit to the pasta factory, a tasting, an interview with the master pasta maker). Length: 120-180 words. A specific email subject line, never “Collaboration”.
Timing tied to the news. Generic pitches get ignored. Pitches hooked to fresh industry news (e.g. a new PGI specification, the Gambero Rosso annual ranking, the Tuttofood trade fair, a viral article on industrial vs artisanal pasta) are 3-5 times more likely to get a reply.
Structured follow-up. First reminder at 5 days, second at 10. After the tenth: you drop that target for 90 days. No stalking.
Measurement. A realistic goal for a new program: 10-15% conversion rate (mention earned per pitch sent). So on 20 well-worked targets a month, 2-3 real mentions.
Sending the pitch to the editor-in-chief of a national publication when your product is hyper-local.
How all of this connects to what I’ve already told you
In my previous articles in this series and in the P1-P4 series I showed you three pieces of the puzzle that fit together today.
I told you about how AI recognizes brand entities in Named Entity Recognition for AI visibility: without a recognizable entity, the mentions you earn aren’t even associated with your brand. I explained in Backlinks as a citation proxy for AI why text mentions matter even without links. And I showed you in Implicit Reference Weight that a mention on an authoritative source, even without a backlink, carries weight in the training of models.
The outreach workflow is the operational activity that turns those principles into a result. Without a process, you know how the machine works but you don’t make it run.
Always keep an up-to-date press kit.
Reverse engineering: 5 artisanal pasta factories, who gets cited by AI and why
I ran a concrete exercise — indicative, not a study — on five Abruzzo artisanal pasta producers of similar size (revenue 1-5 million, 10-30 employees, mixed B2B/B2C/export sales). I’ll call them Pasta Factory A, B, C, D, E for confidentiality.
The test: I asked ChatGPT and Perplexity, with 8 different queries, to recommend “Abruzzo artisanal pasta factories”, “egg pasta Teramo”, “Abruzzo PGI pasta producers”, and similar variants.
The result: two of the five (A and C) were cited in 6-7 answers out of 8. The other three (B, D, E) appeared in 0-2 answers out of 8, often only as a geographic list with no qualitative description.
So I went to look at their outreach work over the last 24 months, as far as visible from the outside (Google News, trade magazine archives, mentions on food blogs). The pattern was clear-cut:
- Pasta Factory A: 14 identifiable text mentions on food publications in 24 months, half with an interview of the owner.
- Pasta Factory C: 9 mentions, many tied to the launch of a PGI line and to trade fair appearances.
- Pasta Factory B, D, E: 0-3 mentions each in 24 months, almost all self-produced (press releases without editorial pickup).
It’s not a perfect correlation — there are other factors (schema markup, Wikidata, site authority) — but the signal is clear: those who do outreach end up in the text passages that models retrieve. Those who don’t remain in an anonymous geographic registry.
Real analysis requires professional mention-monitoring tools and a larger sample. This is an entry-level check to tell you which way to look.
The mistakes I see most often in SME pitches
In my work with artisanal companies — food, wine, shoemakers, ceramics — I see four recurring mistakes.
The self-referential pitch. “Our company, founded in 1952, is a leader in pasta production…”. The journalist closes the email at the second line. The pitch must open with something that interests the publication’s reader, not with your story.
The generalist target. Sending the pitch to the editor-in-chief of a national publication when your product is hyper-local. Better 20 vertical journalists who actually read it than 200 generic addresses pulled from a purchased database.
The lack of ready material. The journalist who replies “send me photos + product sheet + founder story in 200 words” and you take 6 days to prepare it all. The piece runs without you. Always keep an up-to-date press kit.
The aggressive or absent follow-up. Either you push every two days (and get blocked), or you never write again after the first no-reply. The right rhythm is 5-10 days, two reminders maximum.
What can you do in the next 30 days?
I’ll give you a three-step operational path.
- Build the list of 20 targets. Open Google News, search your industry keywords, filter to the last 12 months, note the 20 most active bylines. Save name, publication, topics covered, contact.
- Prepare 3 angles for the pitch. An unreleased internal data point (volumes, percentages of your business), an expert comment on a hot topic, an offer of access (visit/tasting/interview). Write each in 150 words.
- Set the monthly rhythm. On the first Monday of every month you send 20 pitches. Mid-month, first follow-up. End of month, second follow-up and closing. You measure mentions earned / pitches sent. Goal: 10-15% by the third month.
It’s not a magic factor. It’s not enough on its own. But it’s the link that turns technical preparation (entities, schema, authority) into a concrete signal that AI models pick up when someone asks a question about your industry.
