Gold demand set records in Q3. None of it set a floor

Tuesday's close was not the story. The report that landed three hours before it was.

Spot gold was $3,920.63 at the 20:00 UTC cut, −0.68% on the session, a day that ran $3,895.59 to $3,969.96 (Kitco, Gate). The dollar index added 0.19% to 101.86 and the 10-year yield rose five basis points to 4.96% (CNBC). At 16:00 UTC the World Gold Council published Q3 Gold Demand Trends: record value, record ETF holdings, the fastest official-sector buying since the fourth quarter of 2024.

None of that contradicts October's 6.8% decline. The question to ask of a demand report is not how much was bought, but who is left to buy at this price.

The record is in dollars, not tonnes

Total demand including over-the-counter activity was 1,368.6t, down 1% from 1,383.2t a year earlier. Strip out OTC and the physical market fell 4.5% to 1,225.1t (WGC report).

What rose 22% was the value: $188bn for the quarter, $566bn year to date, both records (WGC release). The reconciling number is the price. Q3 averaged $4,262.20 an ounce against $3,456.50 a year earlier, up 23%. In tonnage this market is flat to slightly smaller. In dollars it set records because gold cost a fifth more — and those dollars are now marked against a spot price 8.0% below that average.

321 tonnes, 120 of them disclosed

Central banks added 321.2t net in Q3 — 42% above Q3 2025's 226.3t, 41% above the five-year average of 227t, and the largest quarter since Q4 2024. Year to date: 675t, up 6% (WGC central banks).

Then the number that belongs beside the headline: only 120t was reported. Unreported activity is about 63% of the council's estimate. Of the disclosed 120t, the People's Bank of China bought 63t and Poland 40t: 104t, or 86% of all reported buying, from two institutions. Kazakhstan and Uzbekistan added 14t each; Russia sold 19t (Kitco, WSJ).

So "central banks are buying" is a two-buyer statement dressed as a market one: if the PBoC pauses, the reported figure halves. And the other 201t is Metals Focus's estimate, built from trade data and surveys — the kind of number that gets revised.

The council calls this a price-sensitive bid, not a floor

The report's most useful sentence is in the outlook, not the data: official purchases "appear sensitive to price", and that relationship is "not sufficiently stable to imply a firm floor for the gold price" (WGC Outlook).

Line that up with the sequence: 57.7t at a $4,872.90 average in Q1, 296.3t at $4,506.30 in Q2, 321.2t at $4,262.20 in Q3. Official demand rises as the price falls — either the dip-buying story the bulls tell, or a statement that reserves managers buy more ounces per dollar when gold is cheaper. Both fit the data; only one puts a bid under a falling market, and the council chose the cautious reading. It also raised its central-bank expectation to at least match 2025 while cutting expected investment growth to +200t from +400t.

The demand that shrank is the demand that buys dips

Jewellery consumption was 312.9t, down 17%. Bar and coin was 267.8t, down 19%. The council's explanation is precise: retail investors "held back amid uncertainty over the gold price outlook", and bar and coin fell "given the lack of a clear trend in the gold price" (press release).

That is the elasticity nobody prices. The buyer who answers a $200 dip with a physical order is exactly the buyer who stood still while gold chopped between $3,870 and $3,970 through October. Physical demand is conditional — it waits for a trend — and in the month gold fell 6.8% (October ledger), the waiting was the story.

The one buyer that can leave inside a week

ETFs went from an outflow of 47.3t in Q2 to an inflow of 211.2t in Q3 — a 258.5t reversal in one quarter, the largest swing in the report. Holdings reached a record 4,256t, and Q3's inflow was a record $31bn despite higher bond yields and a stronger dollar (WGC ETF data).

The council reads it as fiscal hedging, and September supports that: $10bn arrived in a month the price fell 8.5%. But ETFs are the only demand category that marks to market daily, and 211t bought in a quarter averaging $4,262 is 8% underwater at Tuesday's price. The 4,256t now held would have been worth about $583bn at Q3's average; at $3,920 it is roughly $536bn.

A fund can sell in a session; a central bank cannot unwind a reserve programme in one. The fastest-moving buyer in this report is the one that just made its largest commitment.

What Wednesday to Friday has to trade

• $3,870.17 / $3,875.21 / $3,895.59 — Friday's one-year low, Monday's low, and Tuesday's double floor at $3,895.59 and $3,896.42. Below the first is the end of the Q3 bull case.
• $3,945 — the shelf JOLTS selling broke; today's plan rests its entry there (the signal).
• $3,969.96 – $3,971.15 — three rejection highs in two sessions, no hourly close above; the $3,975.64 and $3,978.15 lows of 23 and 26 October sit just above as the old floor, now the ceiling.
• $3,979 — two hourly closes here reclaim the broken floor.
• $3,827 — 2R, below the double bottom.

Macro is not ambiguous: ten-year at 4.96%, the 10-year real yield at 2.62% (FXEmpire), the dollar at 101.86, and 73-79% odds of a 25bp hike at the 8-9 December FOMC (CNBC, Barchart). Tuesday's JOLTS print, 7.523m openings against 7.15m expected, pushed all four the same way (Kitco). Payrolls land at 13:30 UTC Friday.

• A hold above $3,945 into payrolls makes the report a dip-buying story after all; watch China's next monthly figure and whether Q4 ETF flows stay positive below $3,930.
• A break of $3,895.59 leaves the $3,870 shelf as the only thing between spot and $3,827. Two hourly closes above $3,979 would invalidate the lower read instead.

What would change this read

Three cheap monitors: China's monthly reserve figure — 63t was more than half its year-to-date buying, so a pause removes the strongest pillar; bar and coin turning up — a Q4 rebound would say October was a pause, not an exit; and ETF flows going negative, which at these prices would show the record 4,256t is momentum wearing a strategic label.

The honest summary is not "demand is near a record". Gold has two buyer bases on different clocks — a slow official one that buys more as price falls, and a fast fund one that has just bought its largest quarter into an 8% drawdown — with retail buyers who stopped buying in between. The first sets a valuation floor over years. The second sets the price this week, and it has never been tested from the other side.

Not financial advice. Spot XAUUSD at the 20:00 UTC cut on 3 November 2026; levels are for the named session. Verdicts on the track record, level sets at today's levels, the macro file on the news board, official-sector data under central banks.

Sources

• World Gold Council — Q3 2026 Gold Demand Trends: 1,369t, record $566bn YTD value
• World Gold Council — press release: 321t official, 211t ETF, $188bn quarter
• World Gold Council — Central Banks: 120t reported, 63% unreported
• World Gold Council — Outlook: purchases "not sufficiently stable to imply a firm floor"
• World Gold Council — ETF holdings and flows: record $31bn Q3 inflow
• Kitco — demand near record as central bank buying surges 42%
• CNBC — gold eases ahead of midterms; dollar and yields higher
• Barchart — 79% odds of a December hike
• WSJ — central banks accelerated gold buying in Q3