Content Ideation at Scale: A Sourcing System, Not a Brainstorm Meeting

When Duolingo's social team explained how they keep generating ideas at a pace that turned a language app into one of the most recognizable accounts on TikTok, the answer was not a burst of genius. It was a calendar entry. In a September 11, 2024 interview (ecommerceexpo.co.uk), Senior Global Social Media Manager Zaria Parvez described it plainly, per Parvez: "Every Tuesday, we have what we call a 'content idea dump.' We gather ideas, look at what audio or captions we want to use, and think about Duolingo's unique point of view."

That is the whole secret, and it is unglamorous. The teams that ship the most content do not brainstorm harder on the morning they need an idea. They run a recurring ritual that turns ideation into a draw from a stocked queue rather than a generation from a blank page. The Monday-panic brainstorm fails not because the people in the room are uncreative but because they are being asked to invent on demand, with no input, against a deadline. A blank page under time pressure produces recycled ideas. A stocked queue produces choices.

This page documents the sourcing system I use to keep an idea queue full across multiple accounts: a fixed weekly ritual, comment mining, and a customer-shoes exercise, all feeding one tagged queue. The system matters more as distribution tightens, because the cost of a recycled idea is no longer just a flat post, it is a flat post that the feed will not distribute. Instagram Reels reach fell 35 percent year over year per the Metricool 2026 Social Media Study (metricool.com), built on 39,762,999 posts, with CEO Juan Pablo Tejela noting, per Tejela: "creators can't just rely on short-form hype anymore." When reach is that scarce, a thin idea is an expensive idea, and a sourcing system is how you stop shipping thin ideas.

A sourcing system, not a generation system

The reframe that fixes ideation is this: you do not have an idea-generation problem, you have an idea-sourcing problem. Generation is the empty-room model, where a person stares at a blank page and tries to produce something from nothing. Sourcing is the model where you pull from inputs that always have material. The Duolingo content idea dump is a sourcing ritual: the team arrives with audio, captions, and observations already collected, and the meeting is a sift, not a strain.

The first input is the recurring ritual itself. Parvez's Tuesday content idea dump works because it is on the calendar every week, which means ideas accumulate between sessions rather than being conjured in the session. The ritual is a container for material the team gathers all week. The fixed day matters: a ritual that floats gets skipped, and a skipped ritual is a queue that runs dry.

The second input is comment mining, which is the audience telling you what they want. The questions, objections, and reactions in your comments and DMs are unfiltered demand signal, and an idea sourced from a comment arrives with that demand attached. Parvez's team turned this instinct into a strategy, per Parvez (ecommerceexpo.co.uk): "We didn't have the bandwidth or skills for creating TikTok videos at the time, but we thought, why not just comment on other people's videos?" The lesson generalizes past commenting: the audience is constantly telling you what to make, and most teams are not reading it as an idea source.

The third input is the customer-shoes exercise, which Rachel Karten described as her core ideation move. She put it directly in her own writing, per Karten: "I put myself in the shoes of the customer, both physically and digitally." You walk through your buyer's actual day and list the frictions your product touches, because those frictions are the use-case hooks that read as native content rather than as ads. Karten writes Link in Bio (milkkarten.net) to roughly 100,000 in-house social managers, and the customer-shoes exercise is the one she returns to because it sources ideas from a real person rather than from a trend feed. Three inputs, one queue, and the empty page never appears.

Step-by-step: building the sourcing system

What good looks like (a worked sample queue)

The numbers below are a clearly disclosed fictional worked example, calibrated against the Metricool 2026 reach baselines and the Buffer 2026 variance findings. The agency, the account mix, and the queue counts are invented. Treat this as an illustration of the method, not a case study.

Team: a fictional three-person social agency running four accounts (a wellness studio, a CrossFit box, a local coffee roaster, and a B2B accounting SaaS). Target output is six clips per account per week, so twenty-four clips. The old model was a Monday brainstorm that produced maybe eight usable ideas across all four accounts, leaving the team scrambling and recycling for the other sixteen slots.

The new model after a month: the Tuesday ritual, comment mining twice a week, and a monthly customer-shoes pass per account feed a single tagged queue. By the start of each week the queue holds roughly forty live candidates across the four accounts, tagged by brand and source. The weekly draw pulls the top twenty-four, and the team always has a buffer for the weeks the ritual runs thin.

The source log after eight weeks: comment-mined ideas were 30 percent of what entered the queue but 55 percent of the clips that cleared the team's save-per-reach floor. Customer-shoes ideas were the strongest performers on profile visits. Trend-feed ideas filled volume but under-performed both on a per-idea basis. The agency reweighted: more comment-mining time, the trend feed demoted to a format-mechanics reference rather than an idea source. The queue stopped being a panic and became an asset, and the differentiation across the four similar accounts held because every idea carried its brand tag from the moment it was sourced.

Where ideation systems break

Failure mode one: the ritual floats. The weekly idea-sourcing block is scheduled "for whenever we have time," which means the first busy week it gets skipped, and the week after that the queue is empty and the team is back to the Monday panic. The fix is a fixed, inviolable day, the way Parvez's team protects Tuesday. The ritual's power is entirely in its recurrence, and a floating ritual has none.

Failure mode two: ignoring the comments as an idea source. The team treats comments purely as community management to reply to, not as demand signal to mine. They miss that the audience is constantly telling them what to make. The fix is to read comments twice a week specifically for ideas and to log the recurring questions, because a comment-sourced idea arrives with demand attached and tends to out-perform a guessed trend idea.

Failure mode three: filtering for feasibility too early. Someone in the sourcing session kills a strong idea because it "sounds expensive," before anyone has tried to find the cheap version. Parvez named this trap directly, per Parvez (ecommerceexpo.co.uk): "If you start thinking about budget and feasibility too early, you restrict yourself from finding the best ideas." The fix is to source first and filter second, with the feasibility gate living in the production pipeline, not in the idea queue.

Failure mode four: chasing volume over choice. The team measures the system by how many ideas it generates and turns ideation into a contest to produce the biggest pile, most of which is never used. The fix is to measure the queue by reliable supply with real selection, not by raw count. A queue that is deep enough to choose from and tagged well enough to differentiate is doing its job; a queue with two hundred untouched ideas is just inventory.

A counter-perspective worth flagging

A fair objection: for a single brand with a sharp, distinctive voice, a heavy sourcing system can dilute exactly what makes the account work. Some of the best brand accounts run on one or two people with strong taste who post reactively, riding the day's energy, and a formal queue with tagging and ranking can sand the edges off the spontaneity that the audience actually follows them for.

There is real merit there. The sourcing system is built for scale: multiple accounts, high cadence, a team that needs reliable supply. A single account with a distinctive voice and a creator who lives inside the niche may genuinely produce better work reactively, because the spontaneity is the product. Forcing a Tuesday ritual onto that creator can replace a strength (instinct) with a process (queue) that produces more consistent but blander output.

I think the deciding factor is whether ideation is currently your bottleneck. If you are missing slots, recycling ideas, or scrambling every Monday, the system fixes a real problem and the consistency is worth the lost spontaneity. If your account is humming on instinct and you are not short of ideas, do not impose a system to solve a problem you do not have. The sourcing system is a cure for the ideation bottleneck, and a brand without that bottleneck does not need the cure.

Metrics to track for the ideation system

Four metrics, split between system-health metrics (is the queue working) and output metrics (do the sourced ideas perform). Output thresholds are floors for accounts in the 0-to-50K follower band, drawn from the Metricool 2026 and Buffer 2026 baselines.

Queue depth (system-health metric): the count of live candidate ideas in the queue at the start of each week, tagged by brand. The target is deep enough to choose from with a buffer, roughly twice your weekly output. Below your weekly output, the queue is empty and you are back to panic. Far above twice your output, you are hoarding ideas you will not use.

Source-to-performance ratio (system-health metric): for each input (ritual, comment, customer-shoes, trend), the share of queue entries it produces versus the share of cleared-the-floor clips it produces. This is the metric that tells you where to spend sourcing time. An input that is 30 percent of the queue but 55 percent of the winners deserves more of your week.

Saves per reach (output metric): the percentage of unique viewers who tap save, the closest organic proxy for intent. This reads whether the sourced ideas actually resonate. Floor for consumer accounts in 2026: roughly 0.40 percent, with the strongest comment-mined ideas clearing it by 2x. An idea source whose clips never clear the floor is a source to demote.

Profile visits per reach (output metric): the percentage of viewers who tap through to the profile, the discovery signal. Customer-shoes ideas tend to drive this because they open on a recognizable friction that pulls curiosity toward the brand. Floor: roughly 1.2 percent for consumer accounts. Reading this per source tells you which inputs drive discovery versus which drive on-clip intent.

Where a planning-first tool fits

The sourcing system runs in a recurring calendar block and an idea queue document. The ritual, the comment mining, and the customer-shoes exercise are human judgment work that no tool replaces, because the demand signal in a comment and the friction in a buyer's day are things a person reads, not a model generates. A planning-first tool earns its slot at the format-mechanics input: surfacing which formats are working in your niche so the team can pair a proven mechanic with a comment-mined or customer-shoes idea. Superdirector is one option here (a hand-built scraper feeding the queue, Foreplay, and manual saved-folders serve the same step). It sits upstream of production and turns a selected idea into a script and a shot plan for export. It does not run your Tuesday ritual, read your comments, or make the judgment call about brand fit, because those are the parts of the system that source the substance, and the substance is the part a tool cannot fake.

Disclosure by Bell Chen, founder of Superdirector: the brand-profile and idea-feed features mentioned here are part of the product I build. It is a planning and intelligence tool that sits upstream of production; it does not generate, edit, schedule, or publish video. Benchmarks and quotes are sourced from the named reports and operators cited inline.

Frequently asked questions