Let AI do the reading and the first draft, and keep three decisions with people: which journalist gets the story, which claim leads the pitch, and whether the email goes out. A pitch reads as personal when it cites the journalist's recent work accurately, offers one specific angle in three short lines, and has passed a manager's check.
A typical pitch generator takes the news, the industry and the audience, then returns a draft that could go to anyone. Muck Rack's own guidance on its generator says most journalists see through generic copy and tells users to add editorial oversight. This workflow adds what generators skip: a note on what each journalist really covers, and a review sheet that a manager signs before anything is sent.
What a one-click pitch gets wrong, line by line
Here is the kind of draft a generator returns when you give it a topic and a tone (illustrative, written for this tutorial):
Hi [first name], I hope this email finds you well! I'm reaching out on behalf of [Client], a leading provider of innovative solar solutions. We are excited to announce a groundbreaking new initiative that we think your readers will love. Would you be interested in learning more? I'd be happy to set up a call.
Nothing in it is false, which is the problem: nothing in it can be checked either. It opens with the sender instead of the story. "Leading", "innovative" and "groundbreaking" are claims a journalist cannot verify, so they are discounted on sight. There is no sign the sender has read a single piece by this person. And the ask, "learning more", makes a reply expensive, because the journalist must first find out what the news is before deciding whether it matters.
You can rewrite a draft like that by hand, but you would be fixing symptoms. The cause is missing inputs: the journalist's actual recent work, the proof behind the claim, and one clear ask. The four steps below put those inputs in before any drafting happens. They follow one running example: a small solar installer, the agency's client, has analysed its own job records. They cover 150 installations in the past year, and the median wait from signed quote to switch-on fell from 31 days to 19. Every figure in this tutorial is invented for illustration.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowStep 1: a beat note for each journalist, from their own pieces
Personalisation fails in one of two ways: it is missing, or it is invented. Ask an AI tool for "a personal line about this journalist" and it will happily praise a piece that does not exist. So the first rule is that the assistant may only work from material you paste in.
Spend about five minutes per journalist collecting their last ten pieces from their author page: headline, date, link and a one-line note of the angle in your own words. Add anything they have published about how they like to be pitched. If you are still building the list itself, media list building with AI covers finding and segmenting the right people before this step. Then run a prompt like this one.
You are helping a PR agency prepare one pitch. Use ONLY the material
between the lines below. Do not use anything you remember about this
journalist or outlet.
----
JOURNALIST: [first name], [outlet], beat as stated on their profile: [beat]
LAST TEN PIECES (headline | date | link | my one-line note of the angle):
1. ...
(paste all ten)
OUR NEWS: [two sentences, figures exactly as in the client fact sheet]
----
Return:
1. The beat in three bullets: topics, usual angle (data, explainer,
profile, opinion), usual sources quoted.
2. The two pieces closest to our news, with date and link, and one line
on why each is relevant.
3. What this journalist has not covered that our news adds.
4. Three things to avoid, based only on the pieces above.
5. Anything you cannot support from the material, marked NOT IN MATERIAL.
Keep each bullet under 25 words.Illustrative output for a trade-press reporter who covers residential energy installations:
- Beat: residential energy installs and installer economics. Usual angle: data-led news and explainers. Usual sources: installers, trade bodies, manufacturers.
- Closest pieces: piece 3 ([date]) on installer waiting times; piece 7 ([date]) on survey backlogs. Both concern the delay between quote and installation.
- Not covered: what an installer can change to shorten that delay.
- Avoid: product-launch framing (none of the ten pieces is a launch); anything that reads as a sales message.
- NOT IN MATERIAL: preferred time of day; whether embargoes are accepted.
What I would change: output like this often adds a judgement the headline does not support, such as calling piece 3 "sceptical", so the adjective comes out. I also open both linked pieces myself. An assistant can quote a date correctly and still attach it to the wrong article, so a person verifies each link before the note is used. That check takes under a minute, and it is the one that stops the invented compliment. Keep the note short and save it in the client's folder next to the preference log described further down.
Step 2: write the hook before the email
A hook is one sentence of 25 words or fewer that says what changed, why this journalist's readers should care, and carries its own proof. Write it before the email, because if you cannot write it there is no pitch yet. Start from the client's approved fact sheet, one page of numbers with the client's permission to publish them, never from what the model remembers. If a release exists, writing a press release with AI that journalists will read covers the source document, and press release drafts, pitch lists and client approvals covers getting the claims signed off.
The prompt for this step is short: give the assistant the fact sheet and the beat note, then ask for three hook lines for this journalist. Each must be 25 words or fewer, lead with the change, the number or the consequence for readers, use no praise adjectives, use only figures that appear in the fact sheet, and end with what you can offer. For the running example, three journalists produce three different hooks from the same data:
| Journalist | Hook line (first draft) | Why it fits |
|---|---|---|
| Trade-press reporter | One installer cut its median wait from signed quote to switch-on from 31 to 19 days across 150 jobs; job-level data available. | Matches her two pieces on installer delays and leads with a figure she can verify. |
| Consumer money writer | Waiting on a solar install? Records from 150 jobs show the median wait fell from 31 to 19 days, plus four questions that predict a delay. | Her readers are buyers, so the data becomes questions they can ask their own installer. |
| Business reporter | A twelve-person installer shortened its median job wait by 12 days without adding crews; the owner will explain the scheduling change. | Fits a small-firm productivity angle and offers a named source who can speak. |
Three tests decide whether a hook survives. The number test: can every figure be traced to a line in the client's file? The swap test: if you replaced the journalist's name with another, would the hook still make sense? If yes, it is too generic. The permission test: has the client approved this claim for publication? The assistant drafts ten hooks in seconds, but choosing among them and passing the three tests is the work that needs a person who knows the client.
Step 3: the three-line pitch
Journalists skim, so the pitch gets three lines. Line one cites a specific piece by this journalist, with its date, and lands the hook. Line two holds the proof and what you can supply: data, a spokesperson, images. Line three asks for one thing that can be answered with a word. Here is the reviewed version of the trade-press pitch, after the assistant drafted it and a person tightened it:
Subject: 150 installs, median wait cut from 31 to 19 days
Hi [first name],
Your piece on [date] about installer waiting times asked why quotes and switch-on dates drift apart. One installer's records for 150 jobs show the median gap falling from 31 days to 19.
Anonymised job data and a one-page method note are ready, and the owner can talk this week.
Shall I send the data pack?
[your name]
Compare it with the generator draft. That one carried no dates, no figures and a vague ask. This one carries a date, two figures and a single yes-or-no question, and the whole email is about 65 words. There is no flattery in it. The accurate reference to the journalist's own piece does the work that "I loved your recent article" tries and fails to do. Attachments stay off unless the preference log says they are welcome.
The prompt for this step repeats the constraints that matter: exactly three lines, the hook from step 2 in line one, one dated citation taken from the beat note, no adjective that is not in the fact sheet, and no sentence beginning with the sender's name or company. If the output breaks a constraint, change the prompt, not just that email, so the fix applies to the next forty.
Step 4: the manager review sheet
The person who wrote a pitch reads what they meant, not what is on the page. That is why the sheet is completed by someone else. It has nine checks, and each one is a yes or no that takes seconds.
| # | Check | What the reviewer looks at |
|---|---|---|
| 1 | Right journalist | Both cited pieces carry this person's byline, are under 90 days old, and the links open. |
| 2 | Claims | Every number and claim matches a line in the client's approved fact sheet, including its qualifiers. |
| 3 | Hook | The news sits in the first sentence, and the line would make no sense with another journalist's name swapped in. |
| 4 | Length | Three lines, about 70 words, no attachment unless the log allows one. |
| 5 | Ask | One question the journalist can answer in a word. |
| 6 | Tone | No flattery, no pressure phrases, no claim of exclusivity unless it is true. |
| 7 | Preferences | Matches the preference log: channel, timing, embargo rules, do-not-contact flags. |
| 8 | Lawful basis | Journalist is on the assessed list, the suppression list was checked, and there is an easy way to say no. |
| 9 | Sender | Goes from a named person's own address with their own signature. |
Here is the sheet filled in for three pitches from the running example:
| Pitch | Result | What the reviewer found |
|---|---|---|
| Trade-press reporter | Pass | All nine checks clear. Sent the same morning. |
| Consumer money writer | Fail, check 1 | Both cited pieces were by a colleague at the same outlet. The assistant had matched on the outlet's name, not the byline. Beat note redone from the right author page. |
| Business reporter | Fail, check 2 | The hook said the wait fell "without adding crews". The client's fact sheet records one new crew starting in March, so the claim was cut and the pitch now leads with the figures alone. |
Two more failures the sheet is built for. The first is the invented compliment: a draft says "loved your recent piece on battery storage" when no such piece is in the pasted list. It appears whenever a prompt invites a "personal touch". The NOT IN MATERIAL rule above is the fix, and check 1 catches any survivor. The second is the stale peg: a "recent piece" that turns out to be fourteen months old. The 90-day window in check 1 exists for that.
You can also use the assistant on the reviewer's side, as a claims ledger. Paste in the pitch and the fact sheet and ask for a list of every factual claim in the pitch, each matched to a line in the fact sheet or marked NO MATCH. The ledger does not replace the review: the manager still opens the fact sheet and the links. What it does is make the claim in line one impossible to overlook.
Below are a draft pitch and the client's approved fact sheet.
List every factual claim in the pitch (numbers, dates, comparisons,
named pieces). For each claim, quote the fact-sheet line that supports
it, or write NO MATCH. Do not rewrite the pitch. Do not add claims.The reviewer should never be the author. In a five-person agency, pair people and swap. Anything that names a person or makes a comparison with a competitor goes to the account director. Keep the signed sheet in the job folder, because it is also your record if a client later asks who approved a claim.
A preference log that stays small and professional
Journalists tell you how they like to be pitched, usually in a short note on their profile or their outlet's contributor page. Record those preferences once and the assistant can respect them every time. Here is a filled-in log for the three journalists (illustrative):
| Journalist | Preference | Where it came from | Review by |
|---|---|---|---|
| Trade-press reporter | Short text email, no attachments; embargo only with two working days' notice. | Her published pitching note, dated. | Before the next campaign |
| Consumer money writer | Wants a named spokesperson and a link to photos. | Her outlet's contributor guidelines page. | Before the next campaign |
| Business reporter | One email only, no follow-ups. | Her public profile text, dated. | Remove if not reconfirmed in twelve months. |
Three rules keep the log defensible. Record only professional preferences the journalist has made public for the purpose of being pitched, and skip anything personal. Note the source and the date, so a reviewer can check it. And treat a do-not-contact entry as final: check 7 on the sheet stops the send. Put the relevant log row into the prompt for steps 3 and 4, so the draft respects it before a human sees it.
A follow-up cadence that stops on time
Follow-ups are where automated outreach most often turns from helpful to irritating, so write the stopping rule before the first send. The one I would use:
- Day 0: send the three-line pitch.
- Day 3, counted in working days: one follow-up, and only if you have something new, such as a chart, a second source, a deadline or a spokesperson's availability. With nothing new, wait.
- Day 8: a closing note, only when the story has a date attached, such as an event or a report publication. Say plainly that it is the last message.
- After that: stop, and leave the journalist alone for 30 days unless there is new, relevant news.
The preference log overrides this schedule. In the running example the business reporter asked for one email only, so she gets no follow-up at all, however good the new chart is. The assistant is useful here because it writes a follow-up from the new element you give it. Tell it to use only that element, to stay under 40 words, and never to use "just checking in", "bumping this" or "circling back". An illustrative day-3 follow-up for the trade reporter:
Hi [first name], one addition to Tuesday's note on installer waiting times: the installer has now split the 150 jobs by property type, and the longest waits were all jobs that needed a separate survey visit. The one-page chart is linked here. The full data pack is still yours if useful.
Log every send and follow-up on the same sheet, so the day-8 and 30-day rules do not depend on anyone's memory. If you send in volume rather than one at a time, deliverability becomes its own problem, and writing cold emails with AI that stay out of spam covers that side.
Legitimate interest for pitch lists: the GDPR basics
A media list is personal data: names, work email addresses and notes on preferences. If it includes journalists in the EU, the GDPR is likely to apply to it. Of the six lawful bases in the regulation, contract, legal duty, vital interests and public task rarely fit a pitch. That leaves consent and legitimate interests, and agencies that cannot collect consent from every journalist on a list tend to look at the second. Whether it fits your situation is a question for your data-protection adviser, but the shape of the test is worth knowing, because it decides what your sheet needs to record.
Guidelines published in 2024 by the EU's data-protection board (the EDPB) say three conditions must all be met: you are pursuing a legitimate interest, the processing is necessary for it, and the individual's interests and rights do not override yours. They also say you should assess and document this before you start. The guidelines add that the regulation's recitals allow direct marketing to be treated as a legitimate interest, but that does not make every marketing activity acceptable.
Two further points matter for pitching. First, a person can object to direct marketing at any time, and once they do, the regulation says their data may no longer be used for that purpose. A journalist who says "please stop" is therefore a hard stop, which is why check 8 includes a suppression list. Second, separate rules on electronic marketing sit alongside the GDPR, and national rules can differ, so ask your adviser how they treat a pitch to a work address. A short assessment for the running example might read as follows:
| Question | Illustrative answer |
|---|---|
| Purpose: what is the interest? | To offer journalists a story relevant to their published beat, on behalf of a client. |
| Necessity: could we do it without this data? | No. We need the journalist's name, outlet, work email and stated beat. We hold nothing else. |
| Balancing: what would the journalist expect? | They publish a work address or pitching note so they can receive relevant pitches (source noted per person). We send only relevant pitches, at most two messages, and honour any request to stop. |
| Safeguards | Suppression list checked before every send; entries without a reconfirmed source removed after twelve months. |
That is a record, not legal advice. Keep it with the sheet, review it when your list or your tools change, and read what to check in an AI tool's privacy policy before pasting any list into an assistant.
One campaign, 40 pitches: where the hours go
Take a five-person agency running 40 pitches for the solar installer, across a mix of trade, consumer and business titles. These timings are illustrative; time yourself on five pitches by hand before you trust any saving.
| Stage, per pitch | By hand | AI draft plus review |
|---|---|---|
| Beat research, including collecting ten pieces | 15 min | 6 min |
| Hook and pitch drafting | 15 min | 5 min |
| Manager review | 0 min | 3 min |
| Follow-up drafting | 5 min | 2 min |
| Total per pitch | 35 min | 16 min |
| Total for 40 pitches | about 23 hours | about 11 hours |
The saving is about 12 hours, and the review step is already paid for inside it. The tempting mistake is to cut the three minutes of review to claim a bigger saving. The trade is deliberate: you spend a little on checking so that you can spend much less on drafting without sending something wrong under the agency's name.
After the sends, the next question is what happened. Replies, links and mentions belong in the same job folder, and AI media monitoring for small PR agencies covers alerts and reports. Feed what you learn back into the preference log and the hook tests, so the second campaign starts better than the first.
Questions PR teams ask before letting AI draft pitches
Should a pitch say it was drafted with AI?
There is no universal rule to quote, so check the outlet's submission guidelines and your client contract. What matters more is accountability: a named person at your agency has verified every claim and sends the email from their own address. If a journalist asks how it was written, answer plainly.
Can I paste a journalist's articles into an AI tool?
The assistant needs the topic and angle of each piece, not the full text. Paste headlines, dates, links and your own one-line notes. Paywalled articles may carry licence terms that rule out copying them into other tools, so avoid pasting them and check the publisher's terms if you are unsure.
What if a journalist asks me to stop emailing them?
Stop at once and add them to a suppression list that the review sheet checks before every send. Under the GDPR, where it applies, a person's objection to direct marketing is not something you can weigh against your own interests: the data may no longer be used for that purpose.
How many pitches can one reviewer check in an hour?
In the worked example a manager spent about three minutes per pitch once the sheet was familiar, which is roughly 20 an hour. If reviews run much longer, the drafts are arriving with weak inputs: send them back for a better beat note or a tighter fact sheet instead of rewriting them yourself.
Further reads
- AI Link Building Outreach: Personalised Without Spam — Same personalisation problem for link outreach, with its own checks.
- Crisis Communication Plan Template: Draft Holding Statements With AI — Holding statements need the same human sign-off as pitches.
- PR for AI Search: How Agencies Earn Citations in ChatGPT Answers — How coverage you earn can end up cited in AI answers.
- ChatGPT Prompts for a Marketing Agency: Briefs, QA and Client Updates — Prompt patterns for briefs, QA and client updates.
- Which Agency Tasks AI Handles Well, and Which It Doesn't — Which agency jobs suit AI and which still need a person.
- AI Media List Tools: Muck Rack vs Semrush AI PR Toolkit vs ChatGPT — A repeatable test: build one media list three ways, check every name, and score Muck Rack, Semrush AI PR Toolkit and ChatGPT on what you find.
- PR Coverage Report: Build One for Each Client With AI — A coverage report structure, the clippings log columns, AI prompts for the summary, and a sample monthly report with the checks a person must do.
- AI Agents for PR: What They Do and How an Agency Supervises Them — A PR agency supervision method for media lists, monitoring and pitch drafts, with a strict approval-before-send gate.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: GDPR Articles 6 and 21 (full text); EDPB Guidelines 1/2024 on processing based on legitimate interest (version 1.0, adopted 8 October 2024); Muck Rack's guidance on using its AI pitch generator.