BrightEdge referral data shows AI search volume still climbing. Against that, a shorter list of brands per answer means each named slot is worth more than it was six months ago.

This edition covers the B2B Tech vertical. The panel holds the same non-branded prompts constant across two dates six months apart. Movement in the numbers reflects how ChatGPT answers changed. What was asked stayed the same. Usage and answer quality can both climb while the named list gets shorter. Measured in AI Catalyst.

What brand mentions measure, and what they do not

Mentions per answer counts how many distinct brands ChatGPT names in a single response, averaged across the tracked prompt set for a funnel stage. It is not citation volume, and it is not market share. A single answer can name many brands or very few, so the count moves independently of how many sources get cited for the same answer.

Source: BrightEdge AI Catalyst. A fixed panel of tracked non-branded B2B Tech prompts, ChatGPT engine, held constant across 15 March 2026 and 13 September 2026.

Where the list got shorter

Where a prompt sits in the buyer's journey now decides how short the list gets. The panel stayed fixed. The prompts stayed fixed. Only the answers changed. Every figure below covers prompts that held the same intent classification on both dates. On those prompts ChatGPT named 5.36 brands per answer in March against 3.87 in September, a 27.9% decline. The fall concentrates at the top of the funnel.

Figures are the fixed-cohort cut: every prompt that carried the same intent classification on both dates. Source: BrightEdge AI Catalyst. A fixed panel of tracked non-branded B2B Tech prompts, ChatGPT engine, held constant across 15 March 2026 and 13 September 2026.

Data collected

Post Purchase carries too few prompts to read as a trend and is excluded. Transactional is printed above and held out of the conclusions below, for the reason given in its own section.

The questions buyers start with got the most selective

Informational prompts are the broadest questions, such as what a technology is or how it works. They went from naming 5.15 brands per answer to 3.50, a 32.0% decline, the steepest fall of any stage.

Microsoft, Google, and Amazon lost the most ground, down 9.5, 6.7, and 6.4 points. A cluster of developer and data-infrastructure names lost ground alongside them: Apache, dbt, Snowflake, Elastic, and HashiCorp.

IBM gained the most, up 9.9 points. NIST.gov, TechTarget, FBI.gov, Cisco and FTC.gov gained too, so institutional and government sources moved up alongside a publisher. The expected publisher squeeze does not appear in this data.

Consideration became a shorter list

Consideration prompts are where someone is comparing named options, such as accounting software for a small business. They went from naming 8.21 brands per answer to 6.39, a 22.1% decline.

Google, Amazon, GitHub, Trend Micro, Apple, and GitLab lost the most ground, down 9.3, 6.1, 3.7, 3.3, 2.0, and 2.0 points.

Every brand gaining ground here is a single-purpose tool.

Why buying-intent prompts are held back

Transactional prompts are where someone is ready to buy. They moved differently over the same window. A third measurement point shows that movement reversing, so the stage is held out of this edition's conclusions until the full weekly series is in. Both figures are printed above, so the omission is visible.

Two dates can show a direction that a third date takes away. So we waited, and printed the figures anyway.

Everything else in this edition holds at every measurement point checked.

Mentions and citations do not always move together

The relationship between how many brands get named and how many sources get cited is not consistent across the funnel.

In Informational, citations per answer rose from 2.30 to 3.12 even as mentions fell. Fewer brands were named, and each was backed by more sources.

In Consideration, citations fell in step with mentions, from 5.70 to 3.25.

In Transactional, citations stayed close to flat, from 1.92 to 1.84, while mentions rose.

What marketers need to know

1. Fight for the winnable Consideration prompts. Pull Consideration-intent prompts for your category in AI Catalyst or AI Hyper Cube, then check where you already show up. Every brand gaining ground here is a single-purpose tool. Find the use case your product owns, and build content that answers that exact question.
2. Audit where you appear stage by stage. A single visibility score hides which stage you are missing from. Split your tracked prompts by intent and find the stage where you are absent, because being absent from a stage is a different fix from ranking low inside it.
3. Hold your top-of-funnel investment. Awareness answers hold their value. They just got harder to win, and the goal there is now earning a place on a shorter list. Pull awareness prompts for your category in AI Catalyst. Identify one to three where you have a right to win but do not appear, then prioritize those.
4. Google is still the channel to plan around. The canonical, specific content that earns organic visibility is also what gets a brand named in an AI answer. Build it once and it works on both. Re-run this funnel comparison quarterly, because funnel-stage presence keeps moving.

How to check this for your own brand

Path one. Open Mentions, filter to your tracked B2B Tech prompt groups, and split by funnel stage. Compare two dates six months apart to see whether your category shows the same shift. Do it in AI Catalyst.

Path two. In AI Hyper Cube, pull the mentioned-brand list for your category across funnel stages for the cross-engine, brand-anchored view.

One caveat. A category with very few Consideration or Transactional prompts will not show a clean trend either way. This check is most useful where your category has enough tracked prompts at each funnel stage to compare, as B2B Tech does here.

Frequently asked questions

What changed in how ChatGPT answers B2B Tech questions?

Over six months ChatGPT named fewer brands per answer, and the decline concentrated at the top of the funnel. Overall presence fell 27.9%, from 5.36 to 3.87 brands per answer.

Which funnel stage lost the most brand presence?

Informational, the broadest and earliest-stage questions, fell 32.0%, from 5.15 to 3.50 brands per answer.

What happened at the buying-intent stage?

Transactional moved differently and is not reported as a finding here. A third measurement point shows that movement reversing, so the stage is held until the full weekly series is complete.

Does this mean publishers are losing ground to owned content?

The data does not support it. TechTarget gained ground in Informational over the same period, and other major publishers stayed flat. That pattern does not carry a publisher-versus-owned-content story.

Is this specific to ChatGPT?

Yes. This measurement is ChatGPT only. Whether Google AI Overviews and AI Mode follow the same funnel-stage pattern is untested and worth checking separately.

How reliable is the funnel-stage classification?

About 15% of prompts were reclassified into a different intent stage between the two dates. Every figure in this edition uses the fixed cohort instead: only prompts that carried the same classification on both dates. Reclassification therefore cannot explain the result.

Technical methodology

Measurement: mentions per answer, the count of distinct brands named in a single ChatGPT response, averaged across the tracked prompt set for each funnel stage.

Panel: tracked non-branded B2B Tech prompts on the ChatGPT engine, matched exactly across 15 March 2026 and 13 September 2026. Every prompt is present in both periods.

Classification: each prompt is assigned a funnel stage by intent. About 15% of prompts were reclassified between the two dates. The headline figures use the fixed cohort, meaning only prompts that held the same stage on both dates.

Cross-check: the same comparison run across all prompts at each date's current classification gives a 25.1% overall decline, against 27.9% on the fixed cohort. Brand-level movements quoted here are computed on all prompts at each date.

Scope: ChatGPT only. Whether the same funnel-stage pattern holds on Google AI Overviews or AI Mode is untested. Google remains the larger channel and the one to plan around.

Key takeaways

Nothing about the content changed. Reddit changed its robots.txt, and every citation gain built on being crawled went with it. A source that had answered one in six Education questions in ChatGPT stopped appearing.

This series has been tracking Reddit's presence in AI citations closely, including an edition on 13 August covering Reddit's rise in search visibility. What follows is what happened in the two weeks after.

Six domains moved into the space inside a week. Two of them were gone again by week 37. The two weeks after the collapse are the useful part, measured in AI Catalyst.

What citation presence measures, and what it does not

Citation presence is the share of tracked prompts that cite a domain at least once. It is not share of citation volume, and it is not market share. A single AI answer cites several domains, so presence figures across domains add up to more than 100%. Reading them as a pie will produce the wrong conclusion.

Source: BrightEdge AI Catalyst. ChatGPT only, a fixed weekly panel of tracked Education prompts, United States, July to September 2026.

The collapse, week by week

Source: BrightEdge AI Catalyst. ChatGPT only, a fixed weekly panel of tracked Education prompts, United States, July to September 2026.

Presence held steady for six weeks, then fell 81% in a single week and reached zero the week after. The Education panel fell further than the account-wide measure. That is what you would expect in Education, where Reddit already sat near the top of the cited set.

What we analyzed

A fixed weekly panel of tracked Education prompts in ChatGPT, measured every week across the same prompts. Because the panel does not change, week-to-week movement reflects the engine's behavior. What was asked stayed the same.

Data collected

Key finding

reddit.com citation presence in the Education panel fell from 17.2% to 3.3% to zero across two weeks in August 2026. Account-wide, the same measure fell from about 9% to 0.5%. The inflection lands in the same week Reddit's robots.txt change took effect.

The panel is fixed, so this corroborates the mechanism behind the drop. It does not prove it. What the data can show is the timing and the size of the drop, and both match when the robots.txt change took effect.

Reddit did not fall on its own

Source: BrightEdge AI Catalyst. ChatGPT only, a fixed weekly panel of tracked Education prompts, United States, July to September 2026.

Community sources and general commercial publishers receded together. Coursera lost 19 points of presence, Indeed 15 and Forbes 14. Reading the event as a Reddit story alone misses that other sources fell in the same window.

What replaced them was narrower and more institutional. Government sites, a university, and professional associations moved into the top set in the same two weeks.

Not everything that moved in stayed in

Source: BrightEdge AI Catalyst. ChatGPT only, a fixed weekly panel of tracked Education prompts, United States, July to September 2026.

The two biggest immediate movers did not hold. The four that held are government, university and professional-association domains. So are the two that faded. onetonline.org and careeronestop.org are both Department of Labor properties. Some institutional sources held and others did not, so domain type alone does not explain durability. Anyone who rebuilt a strategy on the first week's leaderboard rebuilt it on noise.

One name in the new top set is a trap. bls.gov did not grow at all. It was already third before the collapse at 21.2% presence, and it became first because everything around it fell faster. Rising in a ranking and gaining presence are different events.

This was a ChatGPT-specific event

Source: Promptwatch via Search Engine Journal, 19 August 2026; Forbes and Gizmodo, 20 August 2026.

A one-engine read would have told a marketer their Reddit strategy had failed everywhere. It had not. Google's surfaces declined gradually across the same window while ChatGPT fell off a cliff, which is why the cross-engine view matters before anyone acts.

Reddit carried a trait the model wanted

Reddit's high citation rate became a playbook of its own. Teams seeded threads in order to be cited and treated presence on the platform as the objective. The trait the model actually wanted was first-hand, specific, unpolished answers. That content already existed on Reddit.

When the channel closed, it closed through an access policy. Content quality was not the reason. Everything built on it disappeared inside two weeks, and no warning was available to anyone optimizing for it. Reddit-based visibility was never a durable asset.

Some of the domains that moved in held position through week 37 and some did not. Both of the domains that spiked and faded are Department of Labor properties, so institutional status alone does not explain which sources stayed.

Citation share earned through one platform's willingness to be crawled is rented. Visibility built on a third-party platform is exposed to that platform's access decisions. Build the underlying expertise and canonical content on a domain you control. Run one test against your top cited sources. Does the citation exist because of what you are, or because a platform currently allows it to be crawled.

What marketers need to know

One. Build the trait, not the placement. What the model rewarded on Reddit was first-hand, specific, unpolished detail. Replicating that on a domain you control takes original data, named practitioners, and direct experience. Publish the material only you have.

Two. Quick wins from one platform's indexing behavior are fragile, whoever the platform is. Ask whether your current AI citation gains would survive a crawler-policy change tomorrow, and assume you would get no notice.

Three. Wait for the second week before crowning a replacement. The biggest immediate movers here did not hold, and two institutional domains faded alongside them. One week of leaderboard movement is not a trend.

Four. Google is still the channel to plan around. The canonical, specific content that earns organic visibility is also easier for a model to recommend by name. Build it where no third party's access policy can remove it.

How to check this for your own brand

Path one. In AI Catalyst, open Citations, filter the domain to reddit.com and scope to your tracked prompt groups. Pull the weekly trend. A single snapshot will not show it. You are looking for a steady baseline followed by a sudden break. A slope will not produce that pattern.

Path two. In AI Hyper Cube, pull the cited-domain list for reddit.com across your category for the brand-anchored, cross-engine view. That is where you compare Google AI Overviews and AI Mode against ChatGPT.

Three things to look for. Whether your category's Reddit presence collapsed in the same week or whether this is specific to Education. Which domains gained presence in the same window, and whether they held into a second week or spiked and faded. Whether your brand already has a foothold in whatever filled the gap.

One caveat. A category where Reddit had a low baseline will not show a dramatic story. This check is most useful where Reddit already sat in the top three or top five cited domains, as it did here.

Frequently asked questions

What happened to Reddit citations on ChatGPT?

Reddit's presence in ChatGPT citations fell sharply in the second half of August 2026. Public trackers reported a fall from 3.8% to 0.5% of ChatGPT Search citations account-wide, an 86% relative decline (Source: Promptwatch via Search Engine Journal, 19 August 2026). The change is attributed to Reddit blocking crawler access through robots.txt.

When did it happen?

The break lands in the week of 17 August 2026, with presence reaching zero in the Education panel the following week.

What replaced Reddit in Education specifically?

Government, university and professional-association domains. state.gov, asu.edu, apa.org and aacnnursing.org entered the top set and held position. Two others, onetonline.org and careeronestop.org, spiked hardest at the moment of collapse and then receded.

Is this permanent?

Unknown. The outcome depends on an access decision. Content has nothing to do with it. The measurable point is that a policy change removed a source in two weeks, and that is the risk worth planning around.

Does this mean brands should stop trying to get cited on Reddit?

It means a Reddit citation should not be the objective. What made Reddit useful to the model was specific, first-hand detail, and that kind of detail is harder to manufacture on an owned domain. Replicating it takes original data, named practitioners, and direct experience.

Was Reddit's high citation rate ever a reliable quality signal?

It was a signal about access and content type. Reddit's own authority as a source never entered into it. That distinction is what the collapse made visible.

How can I check whether this happened in my own industry?

Follow the two paths in the section above. Pull the weekly trend in AI Catalyst and the cross-engine view in AI Hyper Cube.

Technical methodology

Measurement: citation presence, the share of tracked prompts citing a domain at least once. This is not share of citation volume. Presence across domains sums past 100% because one answer cites several sources.

Engine: ChatGPT. Market: United States. Period: week 28 to week 37, 2026. Panel: a fixed weekly set of tracked Education prompts, held constant so week-to-week movement reflects engine behavior.

The series is plotted as absolute presence and is never indexed to a baseline week. A short series indexed to one week makes every later movement a function of that week.

Attribution. The panel is fixed and cannot observe cause directly. The inflection lands in the same week Reddit's robots.txt change took effect. That timing supports the mechanism. It does not prove it.

Scope. One category, one engine, one account's tracked panel. Directional for Education. It is not a claim about every category, and the account-wide comparison is included so the reader can see the difference.

Every figure quoted anywhere in this edition is a share or a change in a share. We publish no raw citation counts and no panel sizes.

Key takeaways