Read signals over time · 11 min read · Updated 14 Sep 2026
Competitor Monitoring Prompts to Create a Weekly Digest
A competitor digest fails in one of two directions: it lists everything that happened, or it lists nothing because nobody had time. These prompts produce the short version, ordered by what changes a decision rather than by date, from material you collected during the period.
A prompt cannot monitor a competitor, and knowing that is the useful part
Every list of competitor monitoring prompts is slightly mis-titled, including the entry that brought you here. A model has no memory of what a competitor’s page said last month. It cannot notice a change, because noticing requires two observations separated by time, and a chat window holds one.
What a model does extremely well is the part that comes after: taking a pile of things that happened, working out which three of them matter, and writing the short version. That is genuinely the job people find hardest, because it requires saying that eleven of the fourteen items are not worth anybody’s attention, and almost nobody wants to be the person who cut something.
So the prompt starts where the collecting stops
Read the prompts below as the second half of a process whose first half is somebody or something watching the sources. If that first half is a person checking pages on a Friday, the digest will be as complete as their Friday was. The prompt cannot compensate for a period nobody looked at, and it will not tell you it is working from a partial collection.
Prompts for turning a period of competitor signals into a digest
The first writes the digest. The second is for audiences who do not follow the category and need the consequence spelled out. The third is pointed at your collection rather than at your readers, and it is the one that finds the week you thought was quiet and was actually unchecked.
The weekly or fortnightly pass over everything you collected.
Here is everything I collected on [COMPETITORS] between [START] and [END], each item with its date and origin: [PASTE]. Write a digest of at most 300 words for [AUDIENCE]. Order by what changes a decision, not by date. For each item give: what happened, what it means for us, and what if anything we should do. Items where the honest answer to the third is "nothing" still belong in the digest - say "no action" rather than manufacturing one. End with a line naming what you would have expected to see in this period and did not.
When a digest is being read by people who do not follow the category weekly.
Take the digest you just wrote. For every item, add one sentence answering: what would have to be true for this to matter to us? Then mark each item DECIDE (needs someone to choose something), WATCH (matters only if it continues), or NOTE (no consequence we can see). If most items land on NOTE, say so plainly at the top rather than inflating them. A quiet period reported as a quiet period is more useful than five items dressed up.
Run it against the same collection, before you send anything.
Here is what I collected on [COMPETITORS] between [START] and [END], and here is the list of sources I check: [PASTE]. Tell me which sources produced nothing this period, and for each one say whether silence is normal for that source at that cadence or whether it suggests I did not actually check it. Then name what I would expect to see from a company doing [SUSPECTED ACTIVITY] that is absent here. I am looking for gaps in my collection, not a summary of what I have.
Handing a competitor digest prompt a complete period
Two of these three are obvious and the middle one is the reason most digests slowly lose their readers.
Everything you collected in the period, each item dated and attributed
- Where it comes from
- Whatever you gathered as you went, from pricing pages to job postings to product announcements. How often each competitor source is worth checking runs from daily to annually, and a digest built on one cadence quietly misses the others.
- What good looks like
- One line per item with the date, the competitor, the source it came from, and the raw text rather than your summary of it.
The state you reported last time
- Where it comes from
- Your previous digest, pasted in. This is the input a model cannot supply for itself, because it has no memory of what it told you last period.
- What good looks like
- The last digest in full, so the model can say what is genuinely new rather than repeating an item that has been in three consecutive editions.
Who reads it, and what they decide
- Where it comes from
- One line you write: the audience, and the kind of decision they are in a position to make.
- What good looks like
- Specific enough to exclude things. A digest for a product team and a digest for a sales leader should share very few items.
- Then run it
- Paste the grounding instruction, then the previous digest, then this period's collection, then the digest prompt. The previous digest goes first so the model reads it as context rather than as material to summarise.
- Before the output leaves the building
- Check that every item names a source you can open and a date inside the period, and that nothing in the digest is an item you already sent last time wearing different words.
Paste the raw items rather than your notes on them. A digest written from summaries is a summary of summaries, and every judgement you made while collecting has been silently promoted to fact by the time it reaches a reader who cannot see the original.
Ranking competitor news by decision rather than by date
Date order is the default because it requires no judgement, and it reliably puts the item that mattered underneath four that did not. The instruction to rank by what changes a decision is doing the single most valuable thing in the whole prompt, and it is worth being specific about what it means.
An item changes a decision when somebody would do something differently having read it: a price move that changes what a rep should concede, a launch that makes a battlecard section wrong, a hiring pattern that suggests a competitor is entering a segment you were about to commit to. An item does not change a decision because it is recent, because it concerns your largest competitor, or because it took effort to find. The last one is the hardest to let go of.
Reporting a quiet period as a quiet period
The instinct in a week where nothing happened is to find something, and it is worth resisting for a practical reason rather than a principled one. A reader who learns that every edition contains three items, whatever the week, correctly concludes that the count is fixed and the ranking is decorative. After that, the edition where something genuinely happened arrives looking exactly like the four before it.
What a quiet week should actually say
That last sentence is the one that earns the edition. Absence is a signal too, and it is the one a summariser will never surface unless asked, because nothing in the collection represents it.
A competitor monitoring prompt runs on whatever week you happened to collect
Everything above assumes a collection exists. In practice that is the half that decays, and it decays in a way that is invisible from inside the digest: a period nobody checked produces a short, calm edition that reads exactly like a genuinely quiet week.
The mechanism is ordinary. Somebody owns the Friday check, then a quarter-end lands, then they are on holiday, then the habit is gone and nobody announces it. Two months later the digest is still going out, still well written, and no longer connected to what the competitors are doing. The gaps prompt above exists precisely because that state is undetectable by reading the digest itself.
Being late here does not usually cost you a single deal in a dramatic way. It costs you the thing the digest was for: the team stops treating it as the place where they find out, and starts finding out from a buyer instead, which is the same information arriving after the decision rather than before it.
The watching is what a platform does rather than what a prompt does. Flares tracks competitor pricing, product changes, launches and hiring continuously, so the period arrives already collected and the digest prompt is working from a complete week rather than from whoever remembered to look.
The ranking still needs you. Which of this week’s changes deserves your team’s attention depends on which deals are open and what you are about to ship, and no amount of collection decides that. The prompts above are for that half, and they get better as the half beneath them gets more complete.
The collection behind a competitor monitoring prompt
Flares watches your competitors continuously and gathers the changes a digest is written from.
Discover Flares14-day free trial · 30-second setup
A weekly competitor digest FAQ
Can ChatGPT monitor competitors for me?
Not in the sense the word usually means. It can summarise, compare and prioritise material you put in front of it, and it does all three well. What it cannot do is notice: it has no memory of what a page said last month, so nothing tells it that anything changed. Monitoring is the watching, and the prompt is what you run afterwards.
What should a weekly competitor digest include?
Fewer items than you collected, each with what happened, what it means for you, and what if anything to do. Items where the honest answer to the third is nothing still belong there, marked as such, because a digest that only ever contains action items teaches readers that its author inflates things.
How long should a competitor digest be?
Short enough to be read on a phone between meetings, which in practice is around 300 words. The constraint is doing real work: a word limit forces a ranking, and the ranking is the only part of a digest that a reader cannot do for themselves.
Should a digest be ordered by date or by importance?
By what changes a decision. Date order is the default because it requires no judgement, and it reliably buries the one item that mattered underneath four that did not. If two items are equally consequential, then date order between them is fine.
How do you stop a digest repeating the same item every week?
Paste the previous digest in as context. Without it the model has no way to know an item is not new, and an ongoing situation gets reported as news in three consecutive editions. With it, you can ask directly for what changed since the last edition rather than for a summary of the period.
What do you do in a week when nothing happened?
Send a digest saying nothing happened, and name the sources that were quiet. A no-change report is genuine information, it takes ten seconds to read, and it keeps the cadence intact. The alternative teaches readers that every edition contains padding, which costs you the attention you will want in the week something does happen.
How many competitors can one digest cover?
As many as you track, provided the ranking is by consequence rather than by competitor. A digest sectioned per competitor forces an item into existence for each one, which is the structural reason those editions get long and stop being read.
Who should receive a competitive digest?
People who can act on it, segmented by what they can act on, which usually means at least a sales edition and a product edition rather than one for everybody. Underneath the weekly note there is usually a longer competitive intelligence report on a quarterly cycle, and the two should not be carrying the same items.
How do you know whether the digest is working?
Not by open rate, which mostly measures the subject line. The useful signal is whether anything downstream references it: a battlecard updated because of an item, a deal strategy changed, a roadmap conversation citing it. Time to insight is the closest measurable proxy, since a digest that people rely on tends to shorten how long questions take to answer.
Can an AI agent send the digest automatically?
It can assemble and send one on a schedule, and the part that needs specifying is what it is allowed to leave out. An agent reporting every difference it finds recreates exactly the noise problem the digest was meant to solve, so the instruction that matters is the ranking rule rather than the sending.
Competitor monitoring that runs between digests
Flares tracks competitor moves as they happen, so your digest starts from a complete period.
Discover Flares14-day free trial · 30-second setup