What 5,055 ChatGPT conversations reveal about AI visibility
We tested 131 B2B SaaS brands across buyer-style questions to understand when brands appear, whose sources shape the answer, what changes in follow-up conversations, and whether visibility leads to favorable framing.
131
B2B SaaS brands
5,055
buyer-style conversations
18,036
answer citations analyzed
Published July 30, 2026
EXECUTIVE SUMMARY
AI visibility is not one score
Recognition was almost universal when a brand was named, but organic discovery was rare.
The tested brand appeared in 99.9% of brand-prompted conversations, compared with only 12.2% of organic conversations. Category language and decision-stage questions improved the odds of appearing, but neither guaranteed discovery.
Appearing in an answer did not mean that the brand controlled the evidence or the narrative.
In organic answers where the tested brand appeared, competitor-owned sources supplied 43.9% of the citations and third-party sources supplied another 38.6%. Only 17.5% led to the tested brand's own properties.
Additional turns added depth more often than they changed who appeared.
Follow-up questions frequently introduced new citations, but rarely introduced a brand that was absent from the first answer. Visibility was also essentially unrelated to whether the overall brand framing was positive or mixed.
Being present, being discovered, owning the evidence, and being framed favorably are different outcomes. A useful AI visibility analysis has to measure them separately.
SIX FINDINGS
What changed when the question, intent, source context, or turn changed
The following findings use aggregate conversation-level results. Each chart shows the evidence; the adjacent interpretation explains what the pattern means and what it does not prove.
FINDING 01 · DISCOVERY
Being known is not the same as being discovered
Explicitly naming a brand almost guaranteed that it appeared in the answer. Asking about a buyer problem without naming the brand produced a very different result.
Brand appearance changes sharply with question context
Share of conversations in which the tested brand appeared
Brand-prompted
99.9%
1,064 / 1,065 conversations
Competitor-led
32.1%
319 / 995 conversations
Organic
12.2%
364 / 2,995 conversations
| Category | Rate | Appeared / tested |
|---|---|---|
| Brand-prompted | 99.9% | 1,064 / 1,065 |
| Competitor-led | 32.1% | 319 / 995 |
| Organic | 12.2% | 364 / 2,995 |
In 1,065 conversations where the buyer named the tested brand, the brand appeared 1,064 times. That confirms recognition, but it says little about whether the model would introduce the brand without help.
In organic questions, the tested brand appeared in only 364 of 2,995 conversations. Competitor-led questions performed better than organic discovery, but still surfaced the tested brand in fewer than one-third of conversations.
The distinction also held at the brand level: 29 brands had no organic appearances, and 78 of 131 appeared in no more than 10% of their organic test set.
KEY TAKEAWAY
Do not report a single visibility score. Separate brand-prompted recognition, competitor-led consideration, and organic discovery. A high score dominated by brand-aware questions can conceal the fact that buyers who have not already heard of the brand may never encounter it.
Rates are conversation-weighted aggregates. "Organic" means neither the tested brand nor a tracked competitor was named in the opening question.
FINDING 02 · CATEGORY ASSOCIATION
Category language creates a discovery bridge
An organic question did not need to name the brand, but the model usually needed enough category context to connect the buyer's problem with a relevant market.
Category language creates a discovery bridge
Organic brand appearance rate
2.7x higher with category language
| Category | Rate | Appeared / tested |
|---|---|---|
| Fully unbranded | 5.4% | 45 / 833 |
| Category-led | 14.8% | 319 / 2,162 |
Fully unbranded questions produced the lowest organic appearance rate: 45 appearances across 833 conversations. When the question included category language without naming the brand, the appearance rate rose to 14.8%.
That is a 2.7x increase, but it should not be mistaken for high discovery. Even with category context, the tested brand was absent from more than eight out of ten conversations.
The direction was not driven by the largest categories. It also appeared in the balanced 88-brand cohort and in 11 of the 12 business functions.
KEY TAKEAWAY
Build explicit associations between the brand, the category, the buyer's problem, and the use case. The goal is not to repeat short SEO keywords; it is to make the product's role in a buyer's situation unambiguous across category pages, use cases, comparisons, documentation, and external coverage.
Category-led and fully unbranded questions are both organic because neither names the tested brand or a tracked competitor.
FINDING 03 · BUYER INTENT
Visibility rises when the buyer is closer to a decision
The tested brands appeared far more often in vendor-selection and recommendation-seeking questions than in broad educational or diagnostic questions.
Organic visibility rises closer to a buying decision
Organic brand appearance rate by buyer intent
Transactional / Vendor Selection
30.2%
91 / 301 conversations
Comparison
14.6%
115 / 789 conversations
Tactical / How-To
11.7%
76 / 649 conversations
Problem-Aware
10.0%
36 / 361 conversations
Educational
5.2%
12 / 230 conversations
Diagnostic
5.1%
34 / 665 conversations
| Category | Rate | Appeared / tested |
|---|---|---|
| Transactional / Vendor Selection | 30.2% | 91 / 301 |
| Comparison | 14.6% | 115 / 789 |
| Tactical / How-To | 11.7% | 76 / 649 |
| Problem-Aware | 10.0% | 36 / 361 |
| Educational | 5.2% | 12 / 230 |
| Diagnostic | 5.1% | 34 / 665 |
Vendor-selection prompts created the strongest organic opening: the tested brand appeared in 30.2% of those conversations. Comparison questions were a distant second at 14.6%.
Educational and diagnostic prompts were the weakest discovery environments, at approximately 5% each.
The broader response-mode split points in the same direction. Recommendation-seeking prompts surfaced the tested brand in 21.5% of conversations (278 of 1,294), versus 5.1% for informational prompts (86 of 1,701) — 4.2 times as often.
This is descriptive rather than causal. The prompts represent different buyer needs and competitive environments, so the result does not mean that brands should abandon educational content.
KEY TAKEAWAY
Measure visibility by decision stage and buyer intent. Educational material can help an answer, but shortlist visibility also requires strong vendor-selection, comparison, use-case, implementation, and proof content that makes the brand relevant when a buyer is ready to choose.
FINDING 04 · SOURCE AUTHORITY
Appearing in an answer does not mean owning the narrative
Even when the tested brand appeared organically, most supporting citations came from competitor-owned and third-party sources.
Appearance does not mean ownership of the narrative
Answer citations in organic conversations where the tested brand appeared
1,757 answer citations
Competitor-owned
43.9% · 771
Third-party
38.6% · 679
Brand-owned
17.5% · 307
Most missed visibility was not a competitor win
Among 2,631 organic conversations where the tested brand was absent
A tracked competitor appeared
38.2%
1,005 conversations
No tracked competitor appeared
61.8%
1,626 conversations
| Source ownership | Share | Citations |
|---|---|---|
| Competitor-owned | 43.9% | 771 |
| Third-party | 38.6% | 679 |
| Brand-owned | 17.5% | 307 |
The tested brand was organically visible in 364 conversations, and those answers contained 1,757 citations. Only 17.5% of the citations led to the tested brand's own properties. Competitor-owned sources supplied the largest share at 43.9%.
Visibility can therefore be borrowed. The model may mention a brand while relying on a competitor's explanation of the category, integration, workflow, or tradeoff.
At the same time, absence should not automatically be labeled a competitor loss. In 61.8% of organic conversations where the tested brand was missing, no tracked competitor appeared either.
KEY TAKEAWAY
Track citation authority separately from mentions. Strengthen first-party evidence where the model needs specifications, comparisons, implementation detail, and proof; build independent third-party validation; and investigate genuine category or association gaps instead of assuming every missed answer was won by a competitor.
The primary source-mix chart is restricted to answer citations in organic conversations where the tested brand appeared. Across all 18,036 answer citations, the split was 36.0% competitor-owned, 39.1% third-party, and 24.9% brand-owned.
FINDING 05 · MULTI-TURN CONVERSATIONS
Later turns added evidence far more often than new brands
Adaptive follow-ups were useful for comparisons, implementation detail, proof, and citations. They rarely rescued a brand that was absent from the first answer.
Later turns added evidence far more often than new brands
Share of 718 organic multi-turn conversations
77.3%
1-turn · 3,905
20.1%
2-turn · 1,017
2.6%
3-turn · 133
Any new citation appeared after turn 1
49.4%
355 conversations
A competitor first appeared after turn 1
1.8%
13 conversations
The tested brand first appeared after turn 1
0.8%
6 conversations
| Later-turn outcome | Share | Conversations |
|---|---|---|
| Any new citation appeared after turn 1 | 49.4% | 355 |
| A competitor first appeared after turn 1 | 1.8% | 13 |
| The tested brand first appeared after turn 1 | 0.8% | 6 |
Multi-turn testing was valuable, but mainly for depth. Almost half of the 718 organic multi-turn conversations added at least one new citation after the first answer.
New brand discovery was much rarer. Only six conversations introduced the tested brand for the first time after turn one, and only 13 first introduced a tracked competitor later.
The result supports a narrower role for multi-turn analysis: it is most useful for seeing how the model refines evidence, comparisons, implementation guidance, objections, and recommendations after the initial answer.
KEY TAKEAWAY
Include multi-turn tests when the goal is to understand evidence depth, tradeoffs, objections, implementation, and recommendation logic. Do not expect routine follow-up questions to compensate for weak first-answer discovery.
Follow-ups were adaptive rather than randomly assigned. Competitor first-appearance detection used exact, case-sensitive entity matching and may miss aliases.
FINDING 06 · BRAND FRAMING
Higher visibility did not guarantee favorable framing
Brands with positive and mixed overall framing appeared throughout the visibility range. More appearances did not reliably translate into a more favorable narrative.
Higher visibility did not guarantee favorable framing
Per-brand organic appearance rate grouped by overall tone
| Tone | Recommendation | Brands | Avg. organic rate |
|---|---|---|---|
| Mixed | Conditional | 84 | 12.3% |
| Positive | Conditional | 24 | 6.7% |
| Positive | Moderate | 13 | 19.9% |
| Positive | Strong | 7 | 13.2% |
Showing the four largest tone × recommendation groups (128 of 131 brands). 3 brands in smaller mixed + moderate/strong groups are omitted. Correlation between organic visibility and positive tone: r = -0.027.
| Brand id | Tone | Organic rate | Recommendation |
|---|---|---|---|
| brand_001 | mixed | 16.0% | conditional |
| brand_002 | mixed | 19.2% | moderate |
| brand_003 | positive | 13.0% | conditional |
| brand_004 | mixed | 44.4% | conditional |
| brand_005 | mixed | 0.0% | conditional |
| brand_006 | positive | 8.7% | conditional |
| brand_007 | mixed | 7.7% | conditional |
| brand_008 | positive | 3.8% | conditional |
| brand_009 | mixed | 24.0% | conditional |
| brand_010 | mixed | 0.0% | conditional |
| brand_011 | mixed | 30.4% | conditional |
| brand_012 | mixed | 0.0% | conditional |
| brand_013 | mixed | 3.7% | conditional |
| brand_014 | positive | 8.7% | conditional |
| brand_015 | mixed | 63.6% | conditional |
| brand_016 | mixed | 30.4% | conditional |
| brand_017 | mixed | 11.1% | conditional |
| brand_018 | positive | 13.0% | conditional |
| brand_019 | positive | 4.3% | conditional |
| brand_020 | mixed | 8.7% | conditional |
| brand_021 | mixed | 20.8% | conditional |
| brand_022 | mixed | 16.0% | conditional |
| brand_023 | positive | 30.4% | moderate |
| brand_024 | positive | 4.2% | conditional |
| brand_025 | mixed | 33.3% | conditional |
| brand_026 | mixed | 4.3% | conditional |
| brand_027 | mixed | 26.1% | conditional |
| brand_028 | mixed | 25.0% | conditional |
| brand_029 | mixed | 20.8% | conditional |
| brand_030 | positive | 34.8% | moderate |
| brand_031 | positive | 0.0% | conditional |
| brand_032 | mixed | 16.0% | conditional |
| brand_033 | positive | 7.4% | moderate |
| brand_034 | positive | 0.0% | strong |
| brand_035 | positive | 0.0% | conditional |
| brand_036 | mixed | 8.3% | conditional |
| brand_037 | mixed | 4.3% | conditional |
| brand_038 | mixed | 42.3% | conditional |
| brand_039 | positive | 9.5% | conditional |
| brand_040 | mixed | 4.8% | conditional |
| brand_041 | positive | 9.5% | conditional |
| brand_042 | mixed | 7.7% | conditional |
| brand_043 | positive | 28.6% | moderate |
| brand_044 | mixed | 10.0% | conditional |
| brand_045 | mixed | 14.3% | conditional |
| brand_046 | mixed | 19.2% | conditional |
| brand_047 | positive | 0.0% | moderate |
| brand_048 | positive | 0.0% | strong |
| brand_049 | mixed | 4.3% | conditional |
| brand_050 | mixed | 0.0% | conditional |
| brand_051 | mixed | 21.7% | moderate |
| brand_052 | mixed | 0.0% | conditional |
| brand_053 | positive | 4.0% | conditional |
| brand_054 | positive | 14.3% | moderate |
| brand_055 | mixed | 7.7% | conditional |
| brand_056 | mixed | 0.0% | conditional |
| brand_057 | positive | 12.5% | moderate |
| brand_058 | positive | 0.0% | conditional |
| brand_059 | mixed | 15.8% | conditional |
| brand_060 | positive | 36.0% | strong |
| brand_061 | positive | 16.7% | conditional |
| brand_062 | mixed | 15.8% | conditional |
| brand_063 | mixed | 4.8% | conditional |
| brand_064 | mixed | 9.1% | conditional |
| brand_065 | mixed | 10.0% | conditional |
| brand_066 | mixed | 0.0% | conditional |
| brand_067 | mixed | 0.0% | conditional |
| brand_068 | mixed | 4.3% | conditional |
| brand_069 | mixed | 26.9% | conditional |
| brand_070 | mixed | 4.8% | conditional |
| brand_071 | positive | 16.7% | conditional |
| brand_072 | mixed | 39.1% | conditional |
| brand_073 | positive | 4.2% | conditional |
| brand_074 | mixed | 4.5% | conditional |
| brand_075 | mixed | 4.5% | conditional |
| brand_076 | positive | 4.3% | conditional |
| brand_077 | positive | 12.0% | moderate |
| brand_078 | mixed | 0.0% | strong |
| brand_079 | mixed | 0.0% | conditional |
| brand_080 | mixed | 16.0% | conditional |
| brand_081 | mixed | 42.9% | conditional |
| brand_082 | mixed | 4.2% | conditional |
| brand_083 | positive | 40.0% | moderate |
| brand_084 | mixed | 19.2% | conditional |
| brand_085 | mixed | 0.0% | conditional |
| brand_086 | mixed | 7.4% | conditional |
| brand_087 | mixed | 0.0% | conditional |
| brand_088 | mixed | 4.3% | conditional |
| brand_089 | mixed | 4.2% | conditional |
| brand_090 | positive | 38.9% | strong |
| brand_091 | mixed | 21.7% | conditional |
| brand_092 | mixed | 4.2% | conditional |
| brand_093 | mixed | 0.0% | conditional |
| brand_094 | positive | 4.5% | conditional |
| brand_095 | positive | 0.0% | conditional |
| brand_096 | mixed | 0.0% | conditional |
| brand_097 | mixed | 7.7% | conditional |
| brand_098 | mixed | 8.7% | conditional |
| brand_099 | positive | 0.0% | strong |
| brand_100 | positive | 31.8% | moderate |
| brand_101 | positive | 0.0% | conditional |
| brand_102 | mixed | 5.3% | conditional |
| brand_103 | positive | 8.0% | conditional |
| brand_104 | mixed | 0.0% | conditional |
| brand_105 | mixed | 0.0% | conditional |
| brand_106 | positive | 27.3% | moderate |
| brand_107 | mixed | 21.7% | conditional |
| brand_108 | positive | 19.0% | moderate |
| brand_109 | mixed | 4.5% | conditional |
| brand_110 | mixed | 4.8% | conditional |
| brand_111 | mixed | 0.0% | conditional |
| brand_112 | mixed | 4.3% | conditional |
| brand_113 | mixed | 8.7% | conditional |
| brand_114 | mixed | 0.0% | conditional |
| brand_115 | mixed | 41.7% | conditional |
| brand_116 | positive | 16.7% | conditional |
| brand_117 | mixed | 20.8% | conditional |
| brand_118 | positive | 12.5% | strong |
| brand_119 | mixed | 25.0% | conditional |
| brand_120 | mixed | 13.0% | conditional |
| brand_121 | mixed | 7.7% | conditional |
| brand_122 | mixed | 13.0% | conditional |
| brand_123 | mixed | 13.6% | conditional |
| brand_124 | mixed | 27.3% | conditional |
| brand_125 | mixed | 0.0% | conditional |
| brand_126 | mixed | 4.2% | conditional |
| brand_127 | mixed | 9.5% | conditional |
| brand_128 | positive | 5.3% | conditional |
| brand_129 | positive | 0.0% | moderate |
| brand_130 | positive | 4.8% | strong |
| brand_131 | positive | 4.8% | conditional |
The point-biserial correlation between organic visibility and positive rather than mixed overall tone was -0.027: effectively zero in this dataset.
The highest-visibility quartile still contained substantially more mixed than positive brands. Conversely, positive framing and strong recommendation patterns also appeared among brands with low organic visibility.
This does not prove that visibility and framing can never influence one another. It shows that they answer different questions and should not be collapsed into a single score.
KEY TAKEAWAY
Measure whether the brand appears and how the model describes it. Audit positioning, caveats, repeated tradeoffs, recommendation strength, and the language used across personas. Increasing visibility without improving the underlying narrative can amplify a mixed or conditional frame.
Tone and recommendation pattern are aggregate framing classifications. Correlation is descriptive and does not establish causation.
ROBUSTNESS CHECK
The main patterns were not driven by one SaaS category
The core patterns were not driven by one SaaS category
| Pattern checked | Supporting functions |
|---|---|
| Recommendation-seeking > informational | 12 of 12 |
| Competitor-owned citations > brand-owned citations | 12 of 12 |
| Category-led > fully unbranded | 11 of 12 |
Balanced-cohort check
Main-cohort runs with exactly 40 conversations and 5 personas
- 11.9% Overall organic appearance
- 5.6% Fully unbranded appearance
- 14.3% Category-led appearance
- 20.9% Recommendation-seeking appearance
- 5.1% Informational appearance
The magnitude varied by business function, especially in smaller groups, but the direction of the central findings remained consistent. The balanced-cohort check also produced results close to the full dataset.
These checks strengthen confidence that the findings are not simply an artifact of larger brands receiving more questions. They do not turn the study into a universal benchmark for every market, model, or time period.
METHODOLOGY
How the test set was built
The study covered 131 B2B SaaS brands across 12 business functions, including Sales & Revenue, Marketing & Growth, Demo, Content & Visual Communication, AI Engineering, Evaluation & Security, and Customer Success, Support & Retention.
STEP 1
Brand profiling
We built a detailed profile for each brand from 15-20 pages on its website. The profile captured category, positioning, audience, capabilities, competitors, use cases, and proof.
STEP 2
User profiles
Using the brand profile, we created three to five user profiles per brand. Each profile described the user's role rather than job title, the problems they face, and the language they would or would not use when speaking with ChatGPT.
STEP 3
Question generation
We generated 20-40 buyer-style conversational questions per brand. The set included organic, brand-led, and competitor-led contexts and intentionally avoided keyword-first questions.
STEP 4
Adaptive multi-turn conversations
Every conversation began with a buyer-style question. Most ended after the first answer. When the initial response warranted deeper exploration, an adaptive follow-up was added. Conversations were capped at three turns.
Question design examples
Instead of: What's the best software in [category]?
Used: Can you recommend software for [problem the user is facing]?
Instead of: [Brand] vs [competitor]
Used: Between [brand] and [competitor], what would you recommend for [problem the user is facing]?
Conversation context mix
Share of all 5,055 conversations
| Context | Share | Conversations |
|---|---|---|
| Organic | 59.2% | 2,995 |
| Prompted | 21.1% | 1,065 |
| Competitor-led | 19.7% | 995 |
Question mix by buyer intent
Share of all 5,055 conversations
| Intent | Share | Questions |
|---|---|---|
| Comparison | 30.1% | 1,522 |
| Brand-Aware | 21.1% | 1,065 |
| Diagnostic | 13.2% | 665 |
| Tactical / How-To | 12.8% | 649 |
| Transactional / Vendor Selection | 11.1% | 563 |
| Problem-Aware | 7.1% | 361 |
| Educational | 4.5% | 230 |
Question context and bias
Share of all 5,055 conversations
| Bias / context | Share | Questions |
|---|---|---|
| Category-led | 42.8% | 2,162 |
| Brand-led | 21.1% | 1,065 |
| Competitor-led | 19.7% | 995 |
| Unbranded | 16.5% | 833 |
Brands by business function
Number of brands
| Business function | Brands |
|---|---|
| Sales & Revenue | 42 |
| Marketing & Growth | 15 |
| Demo, Content & Visual Communication | 12 |
| AI Engineering, Evaluation & Security | 11 |
| Customer Success, Support & Retention | 10 |
| Developer Infrastructure & Integrations | 10 |
| Voice AI & Conversational Systems | 8 |
| Product Adoption & Digital Experience | 7 |
| Workplace Productivity & Knowledge | 6 |
| AI Agents & Workflow Automation | 4 |
| Product, Feedback & Research | 4 |
| People, Learning & Performance | 2 |
Download the complete research deck
Get the charts, methodology, and six findings in a presentation-ready format.
Download the PDFFind out what AI assistants say about your brand.
Share your website. We’ll test real buyer questions, analyze how AI assistants represent your brand, and send you a reviewed AI Visibility analysis. Every analysis is reviewed by our team before delivery.