Obsurfable“How visible are we in AI search?” collapses into a single percentage on too many dashboards. One number hides engine-specific retrieval, daily churn, and the fact that most brands are invisible on at least one major surface.
The Pineprompt AI Search Visibility Benchmark 2026 (data window September 30–October 13, 2026) is a useful antidote. Researchers ran 200 non-branded US English buyer prompts across 10 categories, scheduled daily on seven engines, and collected 18,792 answers with citations and named brands extracted.
The headline for strategists is not “ChatGPT cites X% of the time.” It is overlap: only 12.9% of the 8,269 brands seen in the study were named by every engine, while 46.2% appeared on only one.
Engines in the study
| Engine | Answers with citations | Avg citations / answer | Avg brands named / answer | Day-to-day brand change |
|---|---|---|---|---|
| ChatGPT | 99.5% | 9.3 | 5.5 | 36.0% |
| Claude (Haiku 4.5) | 95.3% | 4.4 | 6.7 | 46.4% |
| Claude (Haiku 5.5) | 96.4% | 5.0 | 8.5 | 41.1% |
| Copilot | 98.0% | 4.8 | 7.6 | 51.3% |
| Gemini | 90.7% | 6.7 | 6.0 | 55.8% |
| Google AI Mode | 94.8% | 5.9 | 6.5 | 45.6% |
| Google AI Overview | 96.3% | 7.1 | 5.7 | 45.1% |
| Perplexity | 100% | 10.1 | 3.9 | 68.8% |
Several patterns jump out:
- Perplexity cites on virtually every answer and carries the most citations per answer — but names the fewest brands on average (3.9). It is citation-dense and mention-selective.
- Claude Haiku 5.5 names the most brands per answer (8.5) in this sample — a different editorial shape than Perplexity’s tighter lists.
- Day-to-day brand change runs 36% (ChatGPT) to 69% (Perplexity). Treating any single run as ground truth is malpractice.
Pineprompt notes a model transition for Claude mid-collection (Haiku 4.5 until ~October 9, 2026 UTC, then Haiku 5.5) — sensible to compare engines within the same window, not as eternal constants.
Brand overlap: a fragmented landscape
Across engines, Pineprompt reports:
- 12.9% of brands were named by every engine in the benchmark.
- 46.2% were named by only one engine.
- Jaccard similarity between Google AI Overview and ChatGPT named-brand sets: 45.6%.
- Google AI Mode vs ChatGPT: 45.2%.
Roughly half overlap is not “close enough.” It means winning ChatGPT does not win Google, and optimizing for Perplexity’s citation stack may not move Gemini mentions.
That rhymes with Obsurfable’s earlier coverage that only 2.7% of domains are cited by all five engines and Foglift’s Q3 2026 finding that engines do not share citation lists. Pineprompt adds brand-name overlap on buyer prompts with a daily time series — stronger evidence for engine-specific programs.
Citations ≠ brand mentions
The benchmark tracks both citations and named brands. They diverge in practice:
- An answer can cite your docs without naming your product.
- An answer can name you while citing a review site or listicle.
Pineprompt’s averages show multiple brands per answer (roughly 4–8 depending on engine) against 4–10 citations — crowded competitive prose.
Teams should report mention rate and citation rate separately, as Lumar’s October 2026 product split also argues (Presence vs Quality framing). A single blended score hides whether you have a reach problem or a positioning problem.
Volatility: what “36% day-to-day change” means
Pineprompt’s day-to-day brand change metric captures how much the set of named brands shifts between consecutive daily runs on the same prompt schedule.
Interpretation guide:
| Churn band | How to read it |
|---|---|
| ~36–46% | Still volatile; multi-day sampling required |
| ~51–56% | Treat weekly snapshots as noisy |
| ~69% (Perplexity) | Single-day “rank” is almost meaningless |
This is not a Pineprompt quirk. HR92’s October 13, 2026 AI Mode repetition test found one prompt identical on 8/8 runs and another different every time. Volatility is query-dependent and engine-dependent.
Operational rule: run fixed prompt sets several times per week; report frequency and confidence intervals, not position. See AI citation volatility.
Methodology strengths and limits
Strengths
- Buyer-shaped prompts across categories — closer to revenue questions than generic trivia.
- Daily scheduling over 14 days — captures non-determinism honestly.
- Seven engines in one design — rare in public datasets.
- CC BY 4.0 licensing with a cite line — good for secondary analysis.
Limits (from Pineprompt)
- US English fixed sample — other markets may differ.
- Non-branded prompts — branded navigational behavior is underrepresented.
- Engines change without notice — numbers describe a window, not a law of nature.
- Brand stats require ≥2 mentions and segment assignment by October 13, 2026.
Use the benchmark to sanity-check vendor claims, not to replace your category prompt list.
Strategic implications by role
Brand marketers
Pick primary and secondary engines by where buyers actually research (sales call transcripts, win/loss, community channels). Budget engine-specific content and PR — listicles for Google-heavy paths, official product pages where ChatGPT leans vendor-first (Onely’s October 2026 agency study found 57% of ChatGPT citations were agencies’ own sites on “best agencies for X” prompts).
SEO and content leads
Stop shipping one “GEO-optimized” article and declaring victory. Build theme coverage (see Bing Topics Portfolio) and page-type coverage per intent.
Analytics and RevOps
Instrument AI referral where UTMs still work; complement with prompt corpora because many journeys never click. ChatGPT’s mid-October UTM stripping (our write-up) makes answer-level tracking more important, not less.
How Obsurfable fits
Public benchmarks describe ecosystem shape. You still need your prompts, your competitors, and your observation history. Obsurfable stores prompts, answers, brands, and citations in a shared corpus — explore in Explorer, align to your company in the App, and compare mention and citation trends across engines without pretending one score rules them all.
FAQ
Is 12.9% “named by every engine” the same as 2.7% “cited by every engine”?
No. Pineprompt counts brand names in answers; Obsurfable’s 2.7% figure is domain citation overlap across five engines. Both say the same thing directionally: universal wins are rare.
Why does Perplexity name fewer brands but cite more?
Likely editorial format: long citation lists supporting a shorter recommended set. Validate on your category prompts — do not assume global behavior from averages.
Should we drop engines with high churn?
No. High churn means measure more often, not abandon the surface.
Can we license Pineprompt’s raw data?
Check their site for dataset terms; attribution is required under CC BY 4.0.
Bottom line
Pineprompt’s October 2026 benchmark — 18,792 answers, seven engines, 14 days — shows dense citations, multi-brand answers, and material day-to-day churn (up to ~69% on Perplexity). Only 12.9% of brands were named everywhere; 46.2% appeared on a single engine. Engine-agnostic GEO scores cannot capture that fragmentation. Measure per engine, per theme, and over time.