# Prompt 05 — Georgia Tourism Overview (quarterly)

Report type 4 in `../reports.md` ("tourism tracker"). Modelled on the
*structure* of TBC Capital's quarterly "Tourism Sector Overview" (1Q26 issue,
27 May 2026, read for its structure only: summary page, international visitor
trips, income from foreign travel, occupancy and room rate in large branded
hotels). PMC Research Center's Monthly Tourism Update, the other public model,
stopped in 2023.

Worked example: `examples/2026-11-07-tourism-overview.md`, built from
`examples/2026-11-07-tourism-inputs.md`.

## Honest scope: what is and is not publicly reproducible

| Block in the bank report | Public source found | Granularity | Verdict |
|---|---|---|---|
| Visitor trips by type (traveller, visitor, tourist, same-day) | GNTA quarterly workbooks (Ministry of Internal Affairs border data) | quarterly, 9-month | reproducible |
| Top source countries, EU + UK, selected markets | GNTA workbooks (country and "EU" sheets) | quarterly | reproducible |
| Income from foreign travel, total | NBG balance of payments, republished in GNTA's "Economic Indicators" workbook | quarterly | reproducible (via GNTA when nbg.gov.ge is not readable) |
| Income from foreign travel, by country | NBG table on nbg.gov.ge; GNTA's overview PDF prints the top four | quarterly | partly: top four from GNTA; the full table only if nbg.gov.ge is readable |
| Occupancy and room rate in large branded hotels, USD excl. VAT | the bank's own panel | monthly | **not reconstructable**; GNTA quotes STR Global (a different panel, in GEL) |
| Extras the bank report does not carry | Geostat inbound survey (spending, nights, purpose); GNTA entry points (land, air, rail, sea, airports); Geostat hotel survey (annual); Geostat air passengers (quarterly, published late) | various | reproducible, with their own lag |

So the public version is a **quarterly overview built on official border and
balance-of-payments data**, with the branded-hotel block replaced by what is
public and labelled as a different measure.

## How to run it

**Option A — model with code execution.** Download the GNTA and Geostat
workbooks listed in Part 1, parse them, write the input pack, then run Part 2.

**Option B — model without tools.** Open the same pages in a browser, copy the
tables named in Part 1 into a pasted input pack (with each file's name and
"last update" date), then run Part 2.

---

## Part 1 — Gather inputs

1. **Visitor trips** — GNTA, `https://gnta.ge/en/research`: the quarter's
   "International Travel <year> <quarter>" workbook, the year-to-date
   workbook, and the same quarter a year earlier. Record traveller, visitor,
   tourist and same-day trips with y/y; the top 15 source countries; the EU
   sheet (total and its member rows, plus the UK); India, China, Central Asia,
   the Middle East and Gulf rows; and the entry points (land, air, rail, sea,
   each airport, the main land crossings). Check that the earlier quarter's
   figures in the new file equal the original file (revisions).
2. **Travel revenue** — NBG's international travel receipts. Try nbg.gov.ge
   first; if it answers with a bot challenge, stop there (never work around
   it) and use GNTA's "Economic Indicators" workbook, sheet "International
   Tourism Receipts", which names NBG as its source. Record each quarter of
   this year and last, the year-to-date sum and y/y. For countries, record the
   USD figures GNTA's quarterly overview PDF prints and nothing read off a
   chart unless the text confirms it.
3. **Visitor spending and stay** — Geostat inbound tourism survey workbooks
   (expenditure by category, nights, visits, main purpose). Geostat prints
   monthly averages for the quarter: say so, and mark any ×3 figure as
   derived. Cross-check Geostat visits against GNTA visitor trips.
4. **Hotels** — what GNTA's overview prints from STR Global (occupancy, ADR,
   RevPAR, by city) and AirDNA (listings, occupancy), labelled with those
   sources; Geostat's annual hotel survey (latest year).
5. **Air passengers** — Geostat "Airports in Georgia" workbook (source: Civil
   Aviation Agency). If the quarter is not yet published, say so and use
   GNTA's air-border visitor trips as the proxy. Never print a figure read
   from a secondary article as if it were primary.
6. **Reference report** — note only its date and title. Never read its
   numbers before the overview is written; they are used once, afterwards, in
   the numbers-only comparison.

---

## Part 2 — Write the overview

```
You are writing a quarterly overview of Georgia's inbound tourism for
professional readers (hotel investors, lenders, airlines, the tourism
administration). Use only the INPUT PACK. Every figure must carry its period
and source; label every derived figure "derived" and every figure that comes
through a second publisher with both names (e.g. "NBG, via GNTA").

TITLE: "Georgia Tourism Overview — <Quarter Year>, official data as of
<date>"

KEY TAKEAWAYS — 5–7 bullets, each with a number.

1. VISITOR TRIPS — table of trips by type for the quarter and its y/y, then
   the year to date; one paragraph on what grew and what fell.
2. SOURCE MARKETS — top 15 countries with y/y; the EU + UK total and its
   share; the fastest risers and fallers with the reason only if the pack
   gives one and names who gives it.
3. HOW VISITORS ARRIVE — land, air, rail, sea; the airports; the main land
   crossings.
4. TRAVEL REVENUE — quarterly receipts for this year and last, the
   year-to-date sum, y/y; the by-country figures the pack confirms. Then
   visitor spending from the survey, kept separate: say plainly that the
   balance-of-payments receipts (USD) and the survey spending (GEL, visitors
   15+) are different measures, and do not reconcile them.
5. HOTELS — state first that the branded-hotel occupancy and room rate in the
   bank report is the bank's own data and is not reconstructable. Then the
   public figures the pack has, each labelled with its source, and any
   inconsistency in them flagged, not fixed. The annual hotel survey.
6. NOT YET PUBLISHED — every block the pack lists as missing, blocked or
   proprietary, one line each with the reason.
7. SOURCES — Institution — dataset — period — URL — "last update" date —
   date accessed.

Register: dense, institutional; no forecasts, no adjectives without a number.
Define y/y once. Re-check every number against the input pack, then output the
overview only.
```