InfluQuestions and answers
Thirty-one questions about Influ: the twenty sources, the score out of 100, what coverage means, scheduling, alerts and your data.
Influ or just asking an AI chat?
Which should I use?
Both, for different jobs. A chat is better the first time. If you want a campaign explained to you, ask Claude or ChatGPT. They can follow a story anywhere. Influ cannot, and does not try.
Influ is for the question you ask more than once — when this week’s answer has to line up against last week’s.
What does a run give me that a chat answer does not?
It says what it missed. Every run reports its own coverage — for example “12 of 17 sources answered”. A chat never tells you what it failed to find. It sounds equally sure either way.
You can repeat it exactly. Same terms, same window, same sources, saved with a reference code. Run it again next month and the difference is real. Ask a chat twice and you get two essays you cannot subtract.
It builds a history. Every run is kept, so you can see movement rather than one verdict. That is also what makes alerts possible. You cannot put a conversation on a timer.
It counts, it does not remember. Real posts, real links, real numbers from named sources — not a model’s impression of what people are probably saying.
In one line?
A chat explains a narrative. Influ measures whether it is moving.
What is the weakest part today?
Coverage, on anything outside the United States. Some sources simply cannot cover your subject — Congress, OpenFEC and the Federal Register have nothing to say about an Israeli campaign — but they are still counted as sources that did not answer. That pushes the confidence figure down for reasons that have nothing to do with your narrative.
One source, GDELT, currently returns nothing at all for Hebrew searches while working normally in English. Both problems are known and being fixed. Until then, read the coverage number as the worst case rather than the true one.
Doing a run
What happens when I press Run analysis?
Two steps, from one press.
First, an AI reads your narrative and works out what to search for: search terms, slogans, which side it is on, and the opposition framing to watch for.
Then it searches every source you have switched on and scores what it finds. The search terms stay on screen while it works, so you can see what is being matched. If they look wrong, edit them and run it again.
Why does it take a minute or two?
It asks every switched-on source in turn, and some are slow. Reopening a finished check is instant — the report is saved, not rebuilt.
Can I paste a link or a picture instead of typing?
Yes. Paste either straight into the box. A link is fetched and summarised, a picture is described by an AI, and the text lands in the box where you can edit it before running.
Can I start a run without opening the check?
Yes. Run now is the first item in the ⋮ menu on each row of the list.
Reading the result
What is the Landing Score?
One number from 0 to 100 for how far a narrative is actually travelling. It is built from three things: attention (30%), whether people are positive or negative (35%), and whether your own words are being repeated (35%).
A run measured across too little gets no letter at all: below 40% of the sources that could have answered, or fewer than four answering, the number is shown as a measurement and no grade is given.
The weights are published on purpose — a score you cannot argue with is not evidence. Compare it with the same check’s earlier runs. Do not compare it with a different campaign’s.
What does “12 of 17 sources answered” mean?
A run can only measure what replied. A source can be switched off, missing a password, rate-limited, or simply have nothing on your subject. A reading taken from fewer places is a weaker reading, so Influ tells you instead of hiding it.
A quiet-looking result from a half-blind run is worse than no result at all.
What are the marks next to each check in the list?
On the left: the grade from the most recent run and how many sources it was measured across. D 5/9 means a weak score, measured on five of the nine sources that could have answered. A dash instead of a letter means the run was not graded: either under 40% of sources answered, or fewer than four did. The score is still shown — what is withheld is the verdict, because a grade drawn from a third of the evidence is not one. A dot means the check has never been scored.
On the right: how many times it has been measured, and a clock if it is on a cadence.
What is the six-character code?
The reference for that check, such as A1B2C3. Quote it if something looks wrong and it identifies exactly which one you mean.
Why did a source return nothing?
Four everyday reasons: no password set up for it, a rate limit, a search term with genuinely no volume, or a platform that does not allow searching. Settings → Data sources shows which are working.
There is a fifth that is easy to mistake for those: the source cannot cover your subject at all. US government records will always be empty for an Israeli campaign. An empty result is not proof that things are quiet.
Sources
Which sources does Influ use?
Twenty. Six that measure attention such as Google searches and Wikipedia views, eight social and news platforms, three US government records, and three betting markets.
You choose which are switched on, and the choice is saved with each check so a later run can be compared fairly.
Then why does it say 17?
Three of the twenty do not produce a result the coverage figure can count. Seventeen is the number a run can actually answer with.
Do they all apply to my subject?
No, and this is a real weakness today. The US government records and Hacker News cannot cover an Israeli or European campaign, but they are still counted against your coverage. Switching them off in Settings → Data sources gives you a truer figure. Counting only the sources that could have answered is the proper fix, and it is not built yet.
What happens if I switch one off?
It is not asked at all. The run is quicker, and the coverage figure is measured against the smaller set rather than pretending nothing changed.
Watching a check over time
How do I get Influ to measure something on its own?
Open the ⋮ menu on any row in the list and pick how often under Re-run this check: every 6 hours, every 12 hours, daily or weekly. That is its cadence.
The Schedule page, in the account menu at the bottom left, lists everything on a timer: when each last ran, when it is next due, and how many runs it has used this month.
Is it really running?
Yes. Influ checks every fifteen minutes for anything due, measures it, saves the run, and raises an alert if something moved past a limit you set.
Two things to know. It only works while Influ itself is running. And a check has to be run once by hand first, because a scheduled run needs search terms to exist. Until then the Schedule page marks it as switched off and says why.
What is kept each time?
Every run, separately: the score, the coverage, the individual numbers, and the report. Nothing overwrites what came before. That history is where the trend line, the alerts and the confidence figure all come from.
Why does a scheduled run not update the report I can open?
Because the two save different things. A scheduled run stores the numbers; the report you open is the full page, and only a run started here builds one. So the newest numbers always come from the newest run, while the page you open is the last one you ran yourself. Running it once by hand brings them back together.
What sets off an alert?
A scheduled run compares the check against its previous run and tells you only what crossed a limit you chose: the score moving, volume changing, mood shifting, a single source spiking, or the other side’s words appearing.
Why did nothing alert when something obviously happened?
Usually the minimum-coverage rule: if too few sources answered, the run is flagged rather than judged. Quiet hours also hold announcements back, though the run still happens.
Plans and billing
What happens if I go over my monthly limit?
Nothing is blocked. The limit is counted and shown, not enforced.
I downgraded. Do my scheduled runs stop?
Not until the end of the period you already paid for. Downgrading stops the next renewal. It does not switch off a cadence that is already running. The Schedule page shows the date it lapses, and afterwards anything your plan does not include is switched off there with the reason given.
Is there billing?
No. Nothing is connected to a payment company. Choosing a plan is a placeholder, and the money figures on the Efforts page are what the plans would have cost, not money anyone was charged.
Your data
Where does it go?
Your checks, search terms and saved runs are stored in a Postgres database in Frankfurt, run by Neon. The app itself runs in Frankfurt too. An earlier answer here said everything stayed in a file on your own machine — that was the laptop build, and it is no longer how Influ runs.
Two things leave those servers: your search terms go to the sources you have switched on, and the narrative you type goes to whichever AI is set up, because that is what turns it into search terms. A scheduled run does the same thing on its own, with nobody at the keyboard.
The Privacy page has the full list.
Who can see what I searched for?
You, and the operator. A Super Admin can see a history of what every account measured and when, along with the IP address and country each measurement was started from — it is how the service is kept running and how abuse gets spotted. Ordinary administrators cannot, and no other account can see your checks at all.
What happens if I delete my account?
It stops working straight away: you are signed out everywhere and cannot sign back in. But your checks and runs are kept, and the operator can restore the account — so closing it is not the same as erasing it. For real erasure, ask, and it is done by hand.
Is feedback private?
Feedback is stored in the same database. An administrator can read and reply to it. A normal account only ever sees its own messages.
Which AI does it use?
Whichever is set up, in order of preference, moving to the next one automatically if one fails. A provider that has just failed is tried last for a while, so a dead key costs a moment rather than a minute.