Get a demo Demo
Brandlight Research Lab · August 2026

The Direct
Line

Three snapshots · 9, 16 and 23 August 2026
Published by ·

Between 9 and 16 August, ChatGPT's share of citations going to third-party pages fell in five unrelated industries, by 2 to 19 percentage points, while a reference engine measured in the same reports held within 1.5 points in every one. The citations that filled the gap went to pages the brand or a competitor publishes directly.

5
Industries
427K
Prompts measured
3
Weekly snapshots
BrandlightResearch
TL;DR

The whole study, in 60 seconds.

Three consecutive weekly snapshots, 9, 16 and 23 August 2026. Every cut below filters to ChatGPT alone and compares it against a reference engine measured in the same reports, the same weeks.

5/5
Industries where ChatGPT's third-party citation share fell in the same week its total citations rebounded, while the reference engine held flat.
−19.4pp
The steepest drop, in health insurance, from 69.7% to 50.3% of ChatGPT's citations in one week. Banking was close behind at −17.7 points.
+20.8pp
Banking's competitor-page citation share, the largest gain in the panel, while its reference engine moved 0.1 points the same week.
0.8pp
The average move of the reference engine across all five industries, against ChatGPT's average of 11.0 points on third-party share alone.
0/5
Industries where the drop reversed the following week. All five held or continued through 23 August.
149.9%
ChatGPT's total citation growth in health insurance the same week, the rebound we flagged in the previous study. This is what filled it.
The Panel

What was measured, and on which dates

The dates, the market and the scale behind every number in this report.

This study compares three consecutive weekly snapshots, 9, 16 and 23 August 2026, across roughly 427,000 measured prompts in the United States, in five industries: health insurance, consumer packaged goods, banking, veterinary diagnostics and telecommunications.

Every cut in this report filters to a single engine, ChatGPT, and states its citations by source owner: third-party, competitor and owned. It is measured against a reference engine in the same reports, over the same weeks, a version-pinned model in most of the panel and two additional live engines where a pinned model was not available. The reference engine cannot explain a ChatGPT-specific movement, so its own change is the drift floor for that industry.

Finding One

Third-party citation share fell in all five industries

In the same week ChatGPT's citations rebounded, the share going to third-party pages fell in every industry measured, while the reference engine moved less than 1.5 points in any of them.

BrandlightResearch

Change in ChatGPT's third-party citation share, by industry

Percentage-point change in the share of ChatGPT's citations going to third-party pages, 9 to 16 August. Zero sits at the right edge, so every bar runs leftward. The bottom row is the reference engine, averaged across the same five industries.

Health insurance69.7% → 50.3% of ChatGPT citations
−19.4pp
Banking60.8% → 43.1%
−17.7pp
Consumer goods70.0% → 57.8%
−12.2pp
Telecommunications67.0% → 63.3%
−3.7pp
Veterinary diagnostics44.8% → 42.6%
−2.2pp
Reference engine, averagedSame reports, same weeks, all five industries
+0.2pp

Every industry moved the same direction, from a 2.2-point drop in veterinary diagnostics to a 19.4-point drop in health insurance. The reference engine's own change ranged from −0.6 to +1.4 points across the same five industries and the same two weeks, which is the scale of measurement drift, not a finding.

One industry moving is a market story. Five unrelated industries, from health insurance to telecommunications, moving the same direction in the same week while a reference engine sits still is a property of the engine.

The Brandlight Take · 01
ChatGPT pulled back from third-party pages in five industries that share no prompts, no brands and no market dynamics, in the same single week. A reference engine sitting a few tenths of a point away confirms it did not happen to the market. It happened to the engine.
Finding Two

It was not a smaller pie, it was a reallocation

ChatGPT's total citations grew in all five industries the same week. Third-party pages grew slower than the total, or fell outright, while competitor and owned pages grew faster.

BrandlightResearch

The full mix, by industry

Percentage-point change in ChatGPT's citation mix, 9 to 16 August, by source owner. The reference column is the same reference engine's move on third-party share, over the same week.

Industry Third-party Competitor Owned Reference engine
Health insurance−19.4pp+19.6pp+0.8pp+1.4pp
Banking−17.7pp+20.8pp−0.2pp−0.4pp
Consumer goods−12.2pp+7.2pp+7.2pp−0.6pp
Telecommunications−3.7pp+3.8pp+1.4pp−0.4pp
Veterinary diagnostics−2.2pp+12.4pp−5.8pp+1.1pp

ChatGPT's total citation count rose in every industry over the same week, from 16.9% in consumer goods to 149.9% in health insurance. A share move during a volume surge can be an artefact of the surge itself, so we checked the absolute counts before calling this a reallocation rather than noise.

In four of the five industries, third-party citations rose too, just slower than the total, so their share fell even as their count grew. Only in consumer goods did the absolute count of third-party citations fall while everything else rose. Competitor-page citations grew faster than the total in every single industry, which a volume surge alone does not produce.

Health insurance's 149.9% citation rebound followed a contraction on the same engine three weeks earlier, which we flagged at the time as a recovery in progress. This is the answer to what that recovery was made of: proportionally more competitor and owned pages than the engine was citing before the contraction, not a return to the old mix at a larger scale.

The Brandlight Take · 02
More citations and a different mix happened in the same week. If the engine had simply cited more of everything, third-party share would have held. It did not, in any of the five industries.
Finding Three

Mostly it went to named competitors

Competitor-page citations rose in every industry in the panel. Owned-page citations rose in three of the five and fell in one, veterinary diagnostics, where the brand's own share fell while its competitors' rose in the same week.

BrandlightResearch

Competitor versus owned, by industry

Same week, same source-owner mix, sorted by the size of the competitor-page gain.

Industry Competitor-page share Owned-page share
Banking+20.8pp−0.2pp
Health insurance+19.6pp+0.8pp
Veterinary diagnostics+12.4pp−5.8pp
Consumer goods+7.2pp+7.2pp
Telecommunications+3.8pp+1.4pp

Only in consumer goods did owned and competitor pages gain by roughly the same amount. In the other four, competitor-page share grew faster, in banking and health insurance by a wide margin. Veterinary diagnostics is the sharpest case: the tracked brand's own citation share fell by 5.8 points in the same week its named competitors gained 12.4.

Being cited alongside a competitor is a different exposure than being cited less. A brand can hold its own citation count steady and still lose ground if the answer now names a competitor's page in the same breath, more often than it did a week earlier.

The Brandlight Take · 03
A citation count that holds steady can still hide a worse week, if the pages gaining ground next to it belong to a named competitor. Owned-page share is not the only number worth watching here.
Finding Four

It held, it did not snap back

A second weekly snapshot, 16 to 23 August, shows the new level holding in all five industries rather than reverting toward where it started.

BrandlightResearch

Third-party citation share, week two

Change from 16 to 23 August, same measure as Finding One. Compare the scale of these bars to the first week's: a snap back toward the old level would show large positive bars here.

Banking43.1% → 41.6%
−1.5pp
Veterinary diagnostics42.6% → 41.3%
−1.3pp
Telecommunications63.3% → 62.4%
−0.9pp
Consumer goods57.8% → 58.0%
+0.2pp
Health insurance50.3% → 51.5%
+1.2pp

None of the five industries reverted toward its 9 August level. Third-party share moved by no more than 1.5 points in either direction in the second week, against moves of 2.2 to 19.4 points in the first. Competitor-page share partly gave back its gain in two industries, health insurance and veterinary diagnostics, while continuing to build in the other three, and owned-page share kept rising in the same two industries where competitor share cooled.

Two consecutive weeks holding is enough to call this a step change rather than a one-week wobble. It is not yet enough to call it permanent.

The Brandlight Take · 04
A one-week move that reverts by the next snapshot is noise. This one did not revert, in any of the five industries, which is what makes it worth reporting rather than waiting on.
What To Do

So what does this mean for my brand?

Three conclusions that follow from the evidence above, and the honest limit of a three-week window.

1. Watch the source-owner mix per engine, not the citation count alone

Total citations rose in all five industries the same week third-party share fell. A dashboard that only reports volume would have read this as a good week everywhere. The mix moved underneath the count, in every industry measured.

2. Give competitor-page citations their own line

In four of the five industries, competitor-page citation share grew faster than owned-page share, and in veterinary diagnostics owned share fell while competitor share rose in the same week. A brand can hold its own citation count and still be losing this comparison.

3. Re-check weekly, on this engine specifically

The reference engine held within 1.5 points across three weeks and five industries. ChatGPT moved by double digits in four of them and held the new level into a second week. A monthly cadence would have caught this a cycle late.

The limit of a three-week window

Three consecutive snapshots establish that this move happened, replicated across five industries, and held for one week after it happened. They do not establish where it settles. Part of what we are measuring is a citation contraction on the same engine resolving into a new mix rather than the old one at a larger scale, and that resolution may not be finished. The next snapshot decides whether the level from 23 August is where this engine's citation mix now sits, or a point on the way somewhere else.

The Brandlight Take · 05
One engine cited fewer third-party pages and more named competitors, in five industries, in one week, and held it into a second. A reference engine three-tenths of a point away confirms it was not the market. It was the engine, and it did not undo itself.