JOURNAL
The Context Problem: What Your AI Doesn't Know About Your Travel Voice
2 — The Voice Problem
5 May 2026 · 7 min read
There's a reason the term "AI slop" exists. It describes content that is technically correct, fluently written, and completely without character — the output of a language model that has no information about who asked for it or why.
Travel writing is especially vulnerable to this. The destinations are all well-documented. The landmarks are all described. The things to do are all listed somewhere. An AI with no context about you can produce a usable travel guide in two minutes, and it will sound like every other travel guide — because it's drawing from the same pool of generic descriptions that every other guide draws from.
The fix is not a different AI. It's context.
Here's exactly what that means.
What "context" actually means
When a writing agent has context about you, it means it has access to some combination of:
CONTEXT LAYERS
══════════════════════════════════════════════════════════════════
LAYER 1 — VOICE PATTERN (hardest to describe, most valuable)
How you open a piece
Sentence rhythm — long then short, or consistently clipped?
What you observe vs. what you skip
Your relationship to the practical vs. the emotional
Words and phrases you reach for repeatedly
The things you'd never write (your banned phrases)
LAYER 2 — BODY OF WORK (explicit, extractable)
Published posts — all of them, indexed
The structure you use for different post types
The destinations you've covered and how
Internal references you make between pieces
Your evolving opinions about places over time
LAYER 3 — TRIP MATERIAL (specific to the current piece)
Journal logs from this trip
Photos and their context
People, places, specific moments
Things that went wrong
Your emotional state through the trip
LAYER 4 — DESTINATION KNOWLEDGE (general, AI already has)
Geography, climate, logistics
What other people have said about the place
Common routes, common issues, common recommendations
Most AI tools give you Layer 4. Some give you Layer 3 if you paste your notes in. Almost none give you Layers 1 and 2 — because those require reading a body of work, not just a prompt.
The output difference between Layer 4 alone and Layers 1-4 together is the difference between a tourist brochure and a post that sounds like you.
The stateless problem, illustrated
Here's what happens when you ask a stateless AI to write about Ao Nang using only your trip notes:
INPUT: Your notes from one trip to Ao Nang
Generic system prompt ("write conversationally, first person")
OUTPUT: A competent first-person guide to Ao Nang
Accurate information
Generic observations
Nothing that couldn't have been written by someone else
Probably uses "nestled" at least once
Here's what happens when the AI has your full archive:
INPUT: Your notes from this trip to Ao Nang
Your notes from the previous two trips to Ao Nang
All your published Thailand posts
All your published posts across all destinations
OUTPUT: A piece that knows this is your third visit
Observations that compare against your own previous observations
Structure that matches how you handle slow-travel pieces
Opening that matches your rhythm — not a generic "Ao Nang is located in..."
The specific detail from Trip 2 that changed how you see the place
No "nestled"
The difference is not subtle once you've seen both.
Why voice is hard to describe but easy to extract
If you asked most travel bloggers to describe their own voice, they'd struggle. "Conversational" and "honest" are the most common answers — and they're both meaningless as instructions to an AI because every travel blogger thinks they're conversational and honest.
What's actually distinctive is harder to articulate:
- Do you open with a scene or a statement?
- Do you tend to write about how a place made you feel, or what you did there?
- What's your relationship to the practical information — does it get its own sections, or does it weave through the narrative?
- How do you end a piece — with logistics, with emotion, or with a question?
- What's your default sentence length when you're in the flow?
You probably can't answer most of those questions accurately from memory. But an AI that has read fifty of your posts knows the answers, because the answers are all there in the body of work. It doesn't need you to describe your voice. It just needs to have read it.
The import as a shortcut
For a travel blogger with an existing body of work, the fastest path to useful AI context is a full blog import.
The Roami pipeline starts with this: give it your blog URL, it reads your published posts, and the voice profile exists immediately. Not perfectly — the model doesn't know which posts you're most proud of, or which ones you'd prefer it forgets — but well enough to produce output that's genuinely closer to your voice on the first draft.
For a new blogger, this path is slower. The voice profile builds as you write, which means the first few posts through the pipeline will still require more editing. By the fifth or sixth post, the agent has enough material to be genuinely useful. By the twentieth, the first drafts barely need touching.
VOICE PROFILE QUALITY OVER TIME
QUALITY
▲
│ ◆─────────────
│ ◆─────╯
│ ◆────────╯
│ ◆─────────╯
│ ◆─────────╯
│ ◆─────╯
└──────────────────────────────────────────────────────────▶ POSTS
1-2 3-5 6-10 11-20 21-40 40+
↑
"Still needs
heavy editing" ↑
"First drafts are
close to final"
The import collapses the early part of this curve if you already have a body of work. You start somewhere in the middle rather than at the beginning.
What this means for how you write
Two practical implications:
First: The more specific your logs, the better the drafts. An entry that says "Thiksey monastery, good light, cold" gives the AI a time and place. An entry that says "Thiksey at dawn, no tourists, prayer flags moving in the wind, kept thinking about the monk we spoke to yesterday — something he said about returning" gives it emotional material, specific detail, and a thread to pull. Both entries take thirty seconds to write. One produces a much better draft.
Second: Consistency in your own writing compounds. If you always handle practical information the same way, if you have a consistent opening structure, if your endings all land on the same register — the AI picks that up and reproduces it. If your published work is all over the place stylistically, the model has a confused picture of who you are and the output will be inconsistent.
This is actually an argument for developing a consistent style as a blogger — not for the AI's benefit, but because consistency in your own work builds authority with readers. The AI just makes the benefit visible faster.
The gap between "sounds like me" and "is me"
One honest limit: even with full context, there will be moments where the draft produces something technically in your style that you wouldn't have written. A sentence that uses your rhythm but lands on an observation you'd have skipped. A structure that follows your pattern but misses the emotional note you'd have hit.
This is fine. The draft is a starting point, not a finished piece. The gap between "sounds like me" and "is me" is where your editing happens — and it's a much smaller gap than between "sounds like a tourist brochure" and "sounds like me."
The goal of the pipeline is not to replace the writing. It's to replace the blank page.
Join the publish tier waitlist →
Related: Why Your AI Travel Writing Doesn't Sound Like You · From Raw Notes to Published Post: The Full Pipeline
Ready to log your next trip?
Join the waitlist →