Edison Moment · Glossary

Definitions that hold up.

The terms that come up in AI-assisted creative work, marketing, and measurement. Written tight enough to quote, long enough to actually mean something.

AI workflow

3
Human-in-the-loop
Human-in-the-loop is a workflow where a person judges, edits and approves every AI-generated output before it reaches a customer or end user. Human-in-the-loop systems trade some AI speed for quality control. The person makes the final call. They check facts, catch hallucinations, hold the brand voice and finish what the model could not. On Edison Moment, work never goes from the AI straight to the client. An Edison checks every output before the client sees it, and the client approves every delivery.

Related: AI hallucination, Brand voice, Approved reviewer

Brand memory
Brand memory is a lasting store of a brand’s voice rules, visual guidelines, past outputs and feedback that AI tools and the people steering them read before making new work. A prompt lives inside one session. Brand memory lives outside the model and lasts across sessions, briefs and people. Every approved output updates it, so the system learns the brand over time. Edison Moment keeps a brand memory for each brand, and what the client approves, rejects or sends back feeds it. That is why work gets better from job to job.

Related: Brand voice, Voice drift

Prompt engineering
Prompt engineering is the practice of designing, refining and structuring the input given to a language or image model to get a specific kind of output. Good prompts include the role, the task, the constraints, the brand voice and at least one example. They say what to avoid. They improve over many outputs as the writer learns where the model fails. On Edison Moment, the Edison steers the AI tools for you. You pay for the work, not for learning what makes a prompt work.

Related: AI creative brief, Human-in-the-loop

Quality

4
AI hallucination
An AI hallucination is a confidently stated but false output from a language or image model, such as a made-up statistic, a product feature that does not exist or an invented quote. Hallucinations are normal behavior for generative models, not edge cases. They happen because the model picks the most plausible next word, not the most accurate one. In marketing copy they carry real cost: legal risk from made-up claims, lost brand trust and search penalties for unsourced content. The fix is process. A person fact-checks every claim before delivery.

Related: Human-in-the-loop, Approved reviewer, Fact-check

Voice drift
Voice drift is the slow shift of AI-generated copy away from a brand’s documented voice over many sessions or long context windows. Drift happens because models have limited context, prompts compound small errors and sampling adds variance. The fix is a lasting brand memory the model reads from, not a prompt that restates the rules each time, plus a person who can spot drift at a glance.

Related: Brand voice, Brand memory, Human-in-the-loop

Approved reviewer
An approved reviewer is a skilled person who has been checked and approved for a specific kind of work before they can deliver it. In a human-in-the-loop system, the person is the constant and the AI model is the variable. The person picks the right model for the brief, prompts it with the brand context loaded and judges the output before delivery. On Edison Moment these people are called Edisons. Each Edison is a marketer, designer, writer or ad specialist approved for that kind of work.

Related: Human-in-the-loop, AI hallucination

Fact-check
A fact-check is the check of every measurable claim in an AI-generated output, such as statistics, citations, product features and competitor specs, against a real source before the output ships. Fact-checking is the most valuable step in making AI-assisted marketing copy. Models state unsourced claims with the same confidence as sourced ones. The person checking either confirms a source or removes the claim. A missed fact-check costs most in finance, health, legal and comparison claims.

Related: AI hallucination, Approved reviewer

Process

7
AI creative brief
An AI creative brief is a structured description of a piece of creative work, written so an AI model and the person steering it can do the work without follow-up questions. A good AI creative brief states the output, the audience, the one thing the work must achieve, the constraint that would make it fail, and the brand voice. It does not set font sizes, exact wording or other production details. The brief shapes the work. Most rework happens because the brief skipped one of these five inputs.

Related: Brand voice, Human-in-the-loop, Prompt engineering

Delivery time
When your work arrives, usually minutes to a few hours. Every quote shows the delivery time before you pay. Small jobs usually arrive in 15 to 30 minutes, medium ones in 1 to 2 hours and large ones in 3 to 6 hours. Edison Pro adds firm terms. Work starts within 15 minutes during Edison Pro hours. On time, or you get back the difference between the Pro price and the standard price. If the Edison Pro part of your job still isn't delivered at twice its promised time (time spent waiting on you doesn't count), you can cancel that part, and anything waiting on it, for a full refund of those parts. The rest of the job carries on. A second Edison Pro reviewer checks the work before you see it. If that check can't happen within an hour, you get the work unchecked and part of the Edison Pro premium back. Each round of changes comes back within 4 hours, or within the job's own delivery time if that's shorter; the clock pauses while we wait on you. If we miss it, you get the Edison Pro premium back. Premium refunds for lateness, late changes and a missed check never add up to more than the premium; a full refund for a very late part is separate.

Related: Fixed-price brief, Two revisions

Two revisions
A two-revisions policy means the price includes two rounds of changes after the first delivery. Any further revisions are billed separately. Two rounds cover most briefs where the first delivery is close but needs targeted changes. Needing more usually means one of two things: a brief that was too vague, or scope creep. Both are best fixed up front, not at revision time. On Edison Moment, each quote shows the revisions included before you pay.

Related: AI creative brief, Fixed-price brief

Incrementality
Incrementality is the share of revenue or conversions that would not have happened without a specific marketing action. It differs from attributed revenue, which counts whatever the attribution model claims. Incrementality is usually measured with geo-split holdouts, on-platform lift studies or quarterly media mix modeling (MMM). A campaign with a strong reported ROAS but no incremental revenue is buying conversions that would have happened anyway. Measuring it is often what gets a CFO to sign off on the next quarter’s budget.

Related: Media mix modeling (MMM), Attribution

Media mix modeling (MMM)
Media mix modeling is a statistical method that estimates how much revenue each marketing channel drives, using past spend, sales and outside factors, without relying on cookies or platform attribution. MMM is one of the most durable ways to measure marketing now that cookies and iOS 14 tracking have faded. It works at the channel level (Meta, Google, TV, OOH), not the user level. Refreshed each quarter, it gives marketing and finance a shared view of what drives revenue. It suits brands with large, steady paid media budgets.

Related: Incrementality, Attribution

Attribution
Attribution is the practice of giving credit for a conversion to one or more marketing touchpoints in the customer journey, using last-click, first-click, linear or model-based rules. Every attribution model is wrong in its own way. The honest view in 2026 is that attribution is broken by cookie loss, walled gardens, iOS 14 and GA4 sampling. The practical workaround is to compare platform-reported data with MMM and incrementality tests. Treat attribution as a tool, not a truth.

Related: Media mix modeling (MMM), Incrementality

CAPI (Conversions API)
CAPI is Meta’s server-side event API. It sends conversion events straight from your server to Meta, which avoids the browser pixel signal lost to ad blockers, iOS 14 and tracking limits. CAPI is no longer optional for a serious Meta Ads program. A pixel-only setup loses a large share of its signal. Setting up CAPI with AEM (Aggregated Event Measurement) and consent mode is standard first work on any Meta account.

Related: Attribution

Creative

2
Brand voice
Brand voice is the consistent set of language patterns, tone, vocabulary and rules that make a brand recognizable in copy, whichever person or tool wrote it. Brand voice is written down in a voice guide, usually a few pages, with dos and don’ts, examples and anti-examples. AI-only tools struggle to hold a voice across sessions because context drifts. Human-in-the-loop work and a lasting brand memory close that gap. Edison Moment carries your brand voice from job to job, so later work is more on-brand than the first.

Related: Brand memory, Voice drift, Human-in-the-loop

Spark Ads
Spark Ads are paid TikTok ads that run from a creator’s own handle instead of the brand’s account, with the brand paying for distribution. Spark Ads often beat brand-handle TikTok ads on CTR and watch time because people see them as creator content, not brand interruptions. The creator has to grant rights through TikTok’s Spark Ads authorization. They work best as part of a TikTok-native paid media plan.

Related:

Pricing

2
Fixed-price brief
A fixed-price brief is a creative job where the price is set before work begins and does not change with the time spent, the revisions included or how hard the work turns out to be. Fixed-price billing removes scope talks mid-project, makes budgets predictable and keeps the provider focused on delivering the spec, not billing hours. The trade-off is a clear scope up front. On Edison Moment, AI scopes each job in chat and shows one fixed price before you pay.

Related: Delivery time, Credit-based subscription, Two revisions

Credit-based subscription
A credit-based subscription is a recurring billing model where a fixed monthly fee buys a budget of credits you can spend across several services, instead of paying for each output. Credit subscriptions can cost less than paying per output when usage is steady, and they stretch across service types. Check the rollover policy, how long credits last and whether a downgrade takes back unused credits. Edison Moment does not use credits. Every job gets one fixed quote before you pay.

Related: Fixed-price brief

More from the Edison Moment product team in the editorial and comparisons. Ready to brief? Start in the catalog.