A few days ago I asked two language models — GLM5.2 and Opus 5 — to design the social media strategy for a newly founded construction firm. The premise was straightforward: zero advertising budget, 100% organic growth, B2B audience in Spain, and one person managing everything. What I received were two documents that perfectly illustrate the difference between "answering well" and "answering well for you."
The brief: exactly the same for both
- Company: Construction Sector.
- Situation: new brand, no audience, no publishable case studies, no marketing team.
- Constraints: zero ad spend, institutional corporate voice, single bilingual profile.
- Goal: first commercial conversation attributable to social media within 90 days.
- Human resource: one person.
GLM5.2: the plan that sounds professional and is impossible to execute
GLM5.2 delivered a 13-section document, impeccably structured, with tables, roadmaps, buyer personas, and a lean editorial calendar. The problem: it was designed for a company that already exists.
It proposed four simultaneous channels from day one (LinkedIn, YouTube, technical blog, and X), at a pace of 3 posts weekly + 1 video monthly + 1 article every two weeks. It recommended maintaining a "purely corporate" voice on institutional profiles, no personal brand. And it defined a generic positioning — "Repairs" — that doesn't differentiate from any other company in the market.
In short: an excellent plan on paper, but one that ignored the most important constraint: there is only one person behind it, that person has no public track record, and time is the scarcest resource.
Opus 5: the assessment that understood the real context
Opus 5 delivered a "CTO's Assessment." Not a marketing plan: an architecture analysis applied to communication. And the first sentence already marked the difference: "The real constraint isn't content: it's that the LinkedIn API doesn't allow the level of automation you probably have in mind."
Its key decisions were right for a one-person company:
- One channel in phase 1: the personal LinkedIn profile (not the company page).
- Three content pillars, not six. Each piece belongs to exactly one.
- Concrete, defensible positioning: "I do hydraulic repairs, fixed-price quotes."
- Strict escalation triggers: don't open a second channel without 8 weeks of cadence, 300 qualified followers, and 3 commercial conversations.
Moreover, Opus 5 verified the actual state of LinkedIn APIs before proposing automation, distinguished between metrics for decision-making and vanity metrics, and set hard lines ("the agent never publishes without your approval").
The difference in one table
| Dimension | GLM5.2 | Opus 5 |
|---|---|---|
| Channels proposed | 4 simultaneous (LinkedIn, YouTube, blog, X) | 1 in phase 1 (personal LinkedIn) |
| Brand voice | Institutional corporate, faceless | Founder's personal profile |
| Positioning | Generic: "Repairs" | Concrete: "Hydraulic installation specialists" |
| Automation | Not detailed | Realistic pipeline with API verification |
| Key metrics | 8 KPIs (likes, impressions, engagement...) | 1 primary: qualified conversations |
| Scaling | Implicit in the plan | Strict, conditional triggers |
| Real constraints | Ignored | Central to every decision |
The lesson for builders working with AI
GLM5.2 wrote a document that looks right. Opus 5 wrote one that is right for this company, at this moment, with these constraints. The difference isn't technical: it's about understanding context.
When you use AI for strategy, don't look for the most complete output. Look for the one that best answers: "Can I execute this by myself tomorrow morning?" If the answer is no, it doesn't matter how professional the document is: it's a plan for a different company.
The first commercial conversation attributable to social media arrived before the 90 days were up.