Tone, length, voice control · Practical AI in Hotel Operations

The single most common AI complaint from hospitality operators is: "the output sounds generic, like every AI-generated email I have ever read." The complaint is correct and the cause is always the same — the prompt did not specify tone, length, and voice with enough precision. Fixing this turns 60%-good output into 92%-good output without changing the underlying model.

Tone specification

Tone is not "professional" or "friendly." Those words are too vague for the model to act on. Tone is described with specific descriptors: "warm but not gushing," "professional but not corporate," "apologetic but not groveling," "confident without being defensive."

Better: anchor the tone to a recognizable reference. "Tone: like a senior reservations manager writing a personal note to a returning VIP." Or: "Tone: like a hotel concierge giving a polite but firm answer to an unreasonable request." The reference does the work that a list of adjectives cannot.

Length specification

Length should be specified in words OR sentences OR paragraphs, never all three. "120 words" is precise. "3 sentences" is precise. "Short" is not. "A brief response" is not.

Length also has a structural component. "3 sentences" produces a different response than "1 paragraph of 3 sentences." The first is unconstrained on paragraph break; the second is explicit. Be explicit when the structure matters (it usually does for guest-facing content).

Voice specification

Voice is the part of the prompt that distinguishes your property's output from every other hotel's output. The voice spec should reference: the property name and category, the audience (international leisure, corporate, returning VIPs), 2-4 specific phrases the property uses, and 2-4 specific phrases the property avoids.

Example: "Voice: a 5-star Antalya resort writing to an international leisure guest. Use 'we' not 'the property.' Use 'your stay' not 'your visit.' Avoid the words 'curated,' 'fresh,' 'experience,' 'journey.' Include one specific reference to the property — sea-facing terraces, our beach club, or the executive lounge."

The compounding effect

Tone + length + voice specifications add 6-8 lines to a prompt template but improve output quality dramatically. The team that invests one afternoon in writing these specifications for their 4-6 most-common workflows saves hundreds of hours over the following year — and produces output that is recognizably theirs, not recognizably AI.