PR Coverage Report: Build One for Each Client With AI

A PR coverage report for a client should contain a short summary, the best pieces with links, a clippings log, an honest reading of reach and tone, and next month's plan. AI can assemble the log and draft the summary from clippings, but a person must check every number and every quote.

The work that eats the time in a coverage report is rarely the thinking. It is the copying: pasting links into a table, finding the outlet's audience figure, writing 23 near-identical lines, and screenshotting each piece. That is the part to give to a tool. What the client pays you for is the reading of it: which pieces mattered, which did not, and what happens next. Keep those two jobs apart in your workflow, and the report gets both faster and more honest.

Follow me on Instagram@sagnikteaches

What the client actually wants from the report

Different clients read the report for different reasons, and the report should lead with what theirs is. A founder wants to know whether the right people saw the right message. A marketing director wants evidence for a budget meeting. A board wants one number and one chart. If you do not know which reader you have, ask them in the onboarding, and write the answer at the top of the report template as the "reader" line.

Connect on LinkedInSagnik Bhattacharya

Most good reports then follow the same shape, in this order:

Subscribe on YouTube@sagnikteaches

1. The summary. Four or five sentences: what was achieved against the month's goal, the best result, one problem and the next step.
2. Highlights. Three to five pieces, with a screenshot and one line on why each matters.
3. The clippings log. Every piece, in a table.
4. Message pick-up. Which of the client's key messages appeared, and where.
5. What the numbers do and do not say. The caveats, in plain words.
6. Next month. Planned pitches and the stories in development.

The clippings log: the columns that make everything else possible

Everything downstream depends on one table, so build it with care. One row per unique piece, never one row per copy.

Where the clippings come from depends on your tooling: alerts, a monitoring service, the client's own notices and your pitching log. For how to set up the alerts side, see the media monitoring tutorial. Whatever the source, deduplicate before you count anything: the same wire story in nine outlets is one story with nine appearances, and the report should say so.

Reach, sentiment and the caveats you owe the client

These three numbers appear in nearly every coverage report, and each has a limit the client should hear from you rather than discover later.

• Reach is an estimate of the outlet's audience, not of readers. A figure of 800,000 for a publication means that is roughly its size, as the publisher or a measurement service states it. It does not mean 800,000 people read your piece. Always name the source of the audience figure and the date, and add the line "estimated audience of the outlet" to the column header.
• Advertising value equivalency is not a result. Multiplying column inches by an advertising rate is a number that looks like money but measures nothing the client earns. AMEC, the measurement body behind the Barcelona Principles (now in their fourth edition, Barcelona Principles 4.0), runs a "Say No to AVEs" campaign asking PR organisations to stop using advertising value equivalency and all its derivatives as metrics, and publishes 22 reasons not to use them. If a client asks for it, offer outcome measures in its place, such as message pick-up and visits from coverage, and send them AMEC's list rather than arguing the point yourself.
• Sentiment from a tool is a guess. Automated tone scoring struggles with irony, with headlines that are positive and bodies that are not, and with quoted criticism inside a favourable piece. Present tone as "positive, neutral, negative or mixed, checked by [name]".

The honest version of the report states these limits in one short box. Clients rarely object to the caveats. They object to finding them out on their own after quoting your numbers to their board.

Where AI fits, step by step

AI is good at five tasks in this workflow: reading a pasted article and filling the log's columns, grouping near-duplicates, spotting which key messages appear, drafting the summary and tidying prose. It is poor at knowing a client's politics, judging whether a piece is important, and recalling facts it was not given.

Step 1: Fill the log from the articles

I will paste an article. Fill these fields and nothing else:
Outlet | Headline | Date | Type (Original / Syndicated / Mention) |
Prominence (Headline / Main subject / Passing mention) |
Messages present (A, B, C from the list below, or None) |
Spokesperson quoted (Yes/No and name) | Tone (Positive/Neutral/Negative/Mixed) |
One-line reason for the tone.
Key messages:
A. The garden centre's spring planting scheme is open to local schools.
B. Every plant is grown within the group's own nurseries.
C. The scheme is free for schools.
Only use what is in the article. If something is not stated, write "Not stated".

Illustrative output for one pasted piece: Outlet: a regional garden magazine; Type: Original; Prominence: Main subject; Messages: A, C; Spokesperson: Yes, the head of marketing; Tone: Positive; Reason: "describes the scheme as welcome and practical". What you fix: the model said Message C was present, but the article said "low-cost", not free. You change it to A only, which in turn changes your message pick-up count. This is the fix to expect most often, so check the message column against the article for every row you intend to report.

Step 2: Deduplicate

Paste the headlines and outlets, not the full text, and ask for groups of pieces that look like the same story. Check the groups by opening two or three links in each, because headlines are rewritten and a tool may group two genuinely different stories.

Step 3: Draft the summary

Below is a checked clippings table for the month, and the client's goal.
Goal: raise awareness of the schools scheme among parents and teachers.
Write a summary of 90 words or fewer for the client, plain English, with:
- the number of unique pieces and what type they are
- the best result and why it mattered to the goal
- one honest weakness
- one next step
Use only numbers that appear in the table. Do not estimate reach. Do not promise outcomes.

The line forbidding invented reach is important. Left alone, a model will often add a total audience figure that is the sum of every outlet's estimate, including the syndicated copies, which overstates it.

Worked example: one month for a garden centre group

Say your agency retains a garden centre group to promote a free schools planting scheme. In March the clippings log looks like this (all figures illustrative):

So the honest headline is "9 unique editorial pieces (8 original and 1 widely syndicated story), plus 6 mentions and social posts", not "23 pieces of coverage". Of the 9 pieces, 6 carried message A, 4 carried message B and 2 carried message C. Five included a quote from the head of marketing. Tone, after a person re-read the pieces, was 7 positive, 1 neutral and 1 mixed; the tool had labelled the mixed one as positive because the headline was favourable while the body questioned the cost to the group.

The "link back" column showed that three pieces linked to the scheme's page. The agency checks the page's analytics for referral visits from those three outlets in the same month and reports the number as it stands, small or not. A small honest number is worth more than a large unverifiable one.

Building the monthly client PDF

You do not need design software. A document template with a fixed layout, exported to PDF, is enough for most retained clients.

1. Page 1: goal line, the summary, three headline numbers (unique pieces, message pick-up, quoted spokespeople) and a "what changed since last month" line.
2. Page 2: highlights, each with a cropped screenshot, the outlet, the date and one line on why it matters.
3. Page 3: the clippings log, shortened to the columns the client wants. Keep the full table as a spreadsheet attachment.
4. Page 4: the caveats box, next month's plan and a request for anything you need from the client, such as an approved quote.

Ask the AI to fill the "why it matters" lines, but give it the client's goal and the message list, so each line is tied to what the client wants, not to generic praise. A line such as "Strong placement" tells the client nothing; "Reaches parents of primary-age children, the scheme's main audience, and carries the free-for-schools message" does. Where this connects to the rest of your reporting stack, see the client reporting tutorial.

Agreeing the measures before month one

Reports go wrong most often because nobody agreed what success meant. Spend twenty minutes at the start of a retainer on three questions, and write the answers at the top of the report template.

1. What is the goal this quarter? Awareness among a named audience, a launch, a reputation repair, recruitment, or support for sales. Each implies different evidence.
2. Who reads the report, and what will they do with it? A board pack needs one page and one chart. A marketing team wants detail it can reuse.
3. Which three outlets, or types of outlet, matter most? Without this, a hundred minor mentions can look like more success than one placement the client truly wanted.

With those answers, the "why it matters" column practically writes itself, and the report can include a plain line such as "two of the three target outlets covered the story this month". The answers also shape the AI prompts: the goal and the target outlets belong in every prompt as context, so summaries are tied to what the client cares about.

Three client types, three different reports

A single template cannot serve every client, and you should resist the temptation to force it. Adjust the emphasis, not the structure.

• A product launch. The report leads on pick-up of the launch message and the quality of the top five pieces. Reach matters less than whether reviewers wrote about the product at all, so list who was sent samples and who responded.
• An ongoing local brand, such as a pet shop or optician. Volume matters less than relevance: local outlets, community groups and the occasional trade piece. The report can be one page, with visits from coverage and any enquiries the client can tie to a mention.
• A reputation issue. The report leads on tone, with the negative pieces first, the response made and any corrections obtained. This is where a person must read every item, and automated tone must not be trusted.

For reputation work, the crisis communication plan tutorial covers the response side, so that the report shows actions as well as clippings.

Presenting a bad month

Some months produce little: a story did not land, a journalist moved on, a competitor dominated the news. The report should still go out on time, and it should say so plainly. A short structure works:

What happened: 3 unique pieces against a monthly aim of 6; none in target outlets.
Why, as far as we can tell: the launch overlapped with a larger industry story,
and two journalists we pitched were on leave.
What we are doing: re-pitching the story with a new angle on the 12th, and
adding two outlets to the target list.
What we need from you: approval for the revised quote by Friday.

Clients tend to trust an agency that explains a weak month and offers a plan more than one that papers over it with a long list of minor mentions. AI is helpful here for drafting a calm, factual first version, but only from facts you have supplied. Ask it not to speculate about causes, and add the causes yourself, because only you know what happened.

Rolling monthly reports into a quarterly view

Every third month, clients often want a trend. Do not ask AI to read twelve months of reports; ask it to work from your clean log. Put each month's counts into a small table and add simple measures:

These figures are illustrative. The point of the table is that every cell can be traced to the log, so when a client asks "where does this come from?" you can answer in a minute. Use AI to describe the trend in one paragraph and to suggest questions for the next quarter, then edit for accuracy. Avoid describing causes you cannot show: "coverage rose as the schools scheme progressed" is a safer sentence than "coverage rose because of the new pitching approach", unless you tracked the pitching.

Linking coverage to something the client can count

Coverage is most persuasive when it connects to an action. You rarely get a clean line from an article to a sale, but you can build a few honest connections that cost almost nothing.

• A dedicated link. Give each major pitch its own short link or tracking parameter on the page you want visited, so referrals from that coverage can be counted in the client's analytics. Check with the client's web team that the parameter will not break anything.
• A named offer or phrase. A scheme name or a code mentioned in the piece lets the client count enquiries that quote it.
• A "how did you hear about us" prompt. One extra line on an enquiry form gives a rough count of people who came via press.
• Branded search. A rise in searches for the client's name in the weeks after major coverage is a soft signal. Present it as that, not as proof.

Even when the numbers are small, such as 40 visits from three articles in a month, report them as they are, with the date range and the source. Over a year, small honest numbers build a record the client can trust, and they make your larger claims believable.

The human check before it goes out

This is the step that protects your reputation. Run through it every month, in the same order.

• Open every link. Confirm it loads, is the piece you think, and has the date in the log.
• Check every quote. Compare quoted words with the article. A model asked to summarise sometimes puts words in a spokesperson's mouth.
• Recount. Recalculate the totals in a spreadsheet from the log rather than trusting the summary.
• Re-read tone for anything the client would be upset to miss. Negative and mixed pieces go in the report, with a line on what you are doing about them.
• Read the report as the client. If a sentence asks "so what?", rewrite it.

The two-million-readers figure nobody could rebuild

Here is how it tends to go (illustrative). A team lets AI draft the summary from the log, and the draft says the coverage "reached over two million readers". The figure comes from adding the outlet audience estimates, including the nine syndicated copies. The client's finance director asks how it was counted, and the agency cannot reproduce it. The repair is to take reach out of the headline, report unique pieces and message pick-up, and move audience estimates to an appendix labelled as outlet audiences. The lesson for your own prompts is to forbid derived totals unless you have calculated them yourself.

Making it faster each month

Once the template exists, the report should take about an hour. Keep the clippings log running through the month rather than building it on the last day, so the monthly job is review, not data entry. Save your three prompts in a shared note, along with the client's goal and message list, so a colleague can produce the report if you are away. Keep a column of "client reaction" in your own notes (which pieces they praised, which sections they skip), and shorten the report accordingly. If your pitching is thin, the best report in the world will not fix it, and the pitch email tutorial addresses the other end of the pipeline. If you would rather map the whole workflow with someone, my AI implementation consultation covers that on a 1:1 call.

Questions PR teams ask about coverage reports

How long should a monthly coverage report be?

Two to four pages for most retained clients: a one-paragraph summary, the best pieces with screenshots, a clippings table, a short note on what the coverage means against the client's goals, and plans for next month. Longer reports tend to go unread, so put detail in an appendix.

Should I report advertising value equivalent?

It is better not to. AMEC, the measurement body behind the Barcelona Principles, campaigns for PR teams to drop advertising value equivalency and all its derivatives as metrics, and publishes 22 reasons not to use them. Report outcomes the client cares about, such as message pick-up, relevant audiences and visits to a page you can track.

Can AI judge sentiment accurately?

It can give a first pass but gets sarcasm, mixed headlines and industry jargon wrong. Treat automated sentiment as a draft and re-read any piece the client would care about. Label the column 'Tone (checked by [name])' so the client sees a person was involved.

Do I need a paid monitoring tool to do this?

No. You can build a clippings log from alerts, a client's own tracking and manual searches, then use AI to organise and summarise it. Paid tools add coverage you would otherwise miss, so decide by how much coverage you are missing, not by features.

Further reads

• Agency Client Onboarding With AI: Questionnaire, Access and Kick-Off — Agreeing the client's goals and reader before reporting starts.
• AI Media Planning for Small Agencies: Channel Mix and Budget Splits — Where earned coverage sits alongside paid and owned channels.
• AI Media List Tools: Muck Rack vs Semrush AI PR Toolkit vs ChatGPT — Choosing a media list tool for building target lists.
• PR for AI Search: How Agencies Earn Citations in ChatGPT Answers — Which coverage AI answers cite, and how to track it.
• Semrush AI Visibility vs Ahrefs Brand Radar: Which Suits an Agency? — Tools that track links and AI visibility from coverage.
• AI for Advertising Agencies: Media Plans and Weekly Budget Checks — Linking earned coverage to paid media plans.
• Media List Building With AI: Find, Check and Segment Journalists — A four-stage media list method: AI finds candidates, you verify beat, recent work and contact data, then segment and keep opt-outs.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: AMEC pages on the Barcelona Principles 4.0 and the Say No to AVEs campaign, read 4 November 2026; sample figures are illustrative.