Fast vs Thinking Mode: Routing Marketing Work by Depth

Fast mode suits marketing tasks with a known answer shape: reformatting, summarising, tagging, short copy variants. Thinking mode suits tasks with trade-offs: positioning choices, budget logic, audience prioritisation, briefs. The skill is not picking a model; it is classifying your work so the router, or your team, sends each task to the right depth.

Why does the fast-versus-thinking split matter to marketers?

GPT-5 was launched by OpenAI as a unified system with a fast default model, a deeper "GPT-5 thinking" model and a real-time router choosing between them based on conversation type, complexity, tool needs and explicit intent. In the API, OpenAI added a reasoning effort setting, including a "minimal" value for quick answers, and a verbosity setting with low, medium and high values to steer answer length. These are small controls with large workflow consequences.

Most marketing teams currently run every task at one depth. Either everything goes through a quick chat window, so strategic questions get shallow answers, or a team discovers deep reasoning and starts using it for everything, so simple jobs become slow and expensive. Neither is a model problem. Both are triage problems.

How do you classify marketing tasks by depth?

At GPT5 Marketing, an independent frontier AI strategy studio, we sort work with three questions:

1. Is there a single correct output? If yes (convert this table, extract these fields, shorten this to 90 characters), go fast.
2. Are there competing constraints? Budget versus reach, brand voice versus platform norms, short-term sales versus long-term equity. If yes, go deep.
3. Would a wrong answer be expensive or hard to notice? If a mistake would quietly shape spend or positioning, go deep and require the model to show its evidence.

A working triage table

• Fast: caption variants from an approved message, social reply drafts, meeting summaries, tagging comments by theme, translating approved copy for review.
• Medium: competitor creative comparisons, content calendar drafts, landing-page critique against a checklist.
• Deep: positioning options, audience prioritisation, budget reallocation logic, creative briefs, measurement plan design, post-mortems.

What goes wrong when depth is mismatched?

Two failure modes recur. The first is shallow strategy: a team asks a fast model "what should our Q4 message be?" and receives a fluent, generic answer that sounds like every other brand in the category. The second is expensive busywork: deep reasoning applied to a task that needed none, adding latency and cost without improving output. The second is annoying; the first is dangerous, because it looks finished.

Budget decisions illustrate this well. Asking a fast model to "optimise our channel mix" invites vanity logic, chasing reported ROAS rather than true profit. A deep pass, given real cost data and an explicit profit definition, will reason through returns, fees and margin. Tools such as Trafix, which focuses on true-profit ROAS for Meta and Google, are a good example of the kind of structured input that makes deep reasoning about spend worthwhile.

What should you feed a deep pass?

Depth without evidence is just longer guessing. Before routing a task to deep reasoning, assemble:

• The frame: problem statement and decision criteria.
• Current audience signal: real posts, tensions and moments in your category. We use SOMIN, an AI audience-research platform, as this layer; its AI agents can pull and structure that signal before the reasoning step.
• Constraints: budget, legal, brand rules, timing.
• A request for counter-arguments and failure conditions.

The Havas Media case study is a useful reference on agency teams putting audience intelligence upstream of planning, which is exactly where deep reasoning needs it.

How should teams set reasoning effort and verbosity in practice?

For teams building internal tools on the API, we suggest defaults by task class rather than by user. Fast tasks run at minimal or low reasoning effort with low verbosity, because the output is consumed by another step or skimmed by a person. Deep tasks run at higher effort with medium verbosity, plus an explicit instruction to list the evidence used. Exploratory tasks, where the team wants to see the model's alternatives, can run at high verbosity, but should be time-boxed.

For teams using chat interfaces, the equivalent discipline is prompting intent clearly. OpenAI notes the router responds to explicit intent, so saying "think carefully about this, compare options against these criteria" is a legitimate lever rather than superstition.

Definitions

• Reasoning effort: an API setting that controls how much a model reasons before answering.
• Verbosity: an API setting that steers the default length of answers.
• Task triage: classifying work by the depth of reasoning it needs before assigning it to a model or a person.

A worked example: one campaign, three depths

Consider a skincare brand planning a monsoon-season campaign in Singapore. The deep pass comes first: given current conversation about humidity, breakouts and routine fatigue, which tension should the campaign own, and what would make each option fail? A strategist picks one. The medium pass follows: turn the chosen tension into a content calendar across three platforms, critiqued against the brand's tone rules. The fast pass closes the loop: generate caption variants from approved messages, tag incoming comments by theme, summarise weekly performance notes.

Notice the order. Deep thinking sets direction once; fast work executes many times. Teams that invert this, generating hundreds of fast assets and hoping strategy emerges, end up with volume and no point of view.

Checklist: is your team routing well?

• Do you have a written list of which tasks go fast and which go deep?
• Does every deep task start with a frame and evidence?
• Do deep outputs cite what they relied on?
• Are fast outputs checked by sampling rather than line by line?
• Does someone review the triage list each quarter?

Routing is the first habit we install in a frontier sprint, because it touches every other workflow. For the broader picture of why this matters to strategy, read what reasoning models change about marketing strategy.

Frequently asked questions

What is the GPT-5 router?

OpenAI describes GPT-5 as a unified system with a real-time router that chooses between a fast model and a deeper thinking model based on complexity, tool needs and explicit intent.

Which marketing tasks need deep reasoning?

Tasks with competing constraints or expensive, hard-to-spot errors: positioning, audience prioritisation, budget logic, creative briefs and measurement plans.

What do reasoning effort and verbosity control?

In the GPT-5 API, reasoning effort sets how much the model thinks before answering (including a minimal setting) and verbosity steers answer length: low, medium or high.