The leaders had both answer presence and a public proof trail.
Profound appeared in 16 of 25 valid answers and Peec AI in 14. HowAICite was not named or cited in any valid answer.
Valid provider answers
Answers naming Profound
Five provider failures excluded
Scope: United States · English · four provider APIs · October 1, 2026, approximately 22:07–22:23 ICT. Requested location handling differed by provider. See collection method.
Profound and Peec AI were the only names present in more than half of valid answers.
Answer incidence counts an answer once per brand. Signal count is the number of exact-name occurrences and can exceed the number of answers.
| Tracked brand | Answers with signal | Answer incidence | Exact-name signals |
|---|---|---|---|
| Profound | 16 / 25 | 64% | 44 |
| Peec AI | 14 / 25 | 56% | 34 |
| Scrunch AI | 5 / 25 | 20% | 12 |
| AthenaHQ | 5 / 25 | 20% | 12 |
| Writesonic | 3 / 25 | 12% | 8 |
| Otterly.AI | 2 / 25 | 8% | 4 |
| ZipTie | 2 / 25 | 8% | 4 |
| Goodie | 1 / 25 | 4% | 3 |
| Evertune | 0 / 25 | 0% | 0 |
| HowAICite | 0 / 25 | 0% | 0 |
These are deterministic exact-name matches in stored answer text, excluding URLs. They do not establish recommendation strength, sentiment, market share, conversion or causality. A zero means no exact match in this panel—not absence from AI search generally.
The visible leaders also publish stable, outcome-led customer evidence.
This is a qualitative review of vendor-owned pages, not verification of the outcomes they report.
A large customer-story library with outcome-led titles across SaaS, finance, healthcare and agencies.
Inspect vendor page ↗A dedicated case-study hub that foregrounds vendor-reported visibility, citation and acquisition outcomes.
Inspect vendor page ↗A detailed Merge story reporting a 7× increase in demo requests from AI search and showing the workflow behind it.
Inspect vendor page ↗A Grüns case study reporting a 6× share-of-voice lift in 60 days, with sequence and timeline sections.
Inspect vendor page ↗A SteelSeries case study built around category rank and a vendor-reported 3.2× conversion outcome.
Inspect vendor page ↗Both maintain public customer or case-study hubs that give crawlers stable entity-to-outcome pages.
Inspect vendor page ↗Inspect second hub ↗Our inference: repeated, crawlable case-study pages create more entity-to-category and entity-to-outcome associations for retrieval systems to encounter. That is a plausible distribution advantage, not proof that case studies caused the answer incidence above.
A fixed question panel with provider failures left visible.
Valid-answer denominators never include timeouts, temporary-unavailable responses or requests rejected by the hourly cap.
| Provider | Model | Attempted | Valid | Failed |
|---|---|---|---|---|
| OpenAI Web Search API | gpt-5.6-terra | 10 | 10 | 0 |
| Perplexity Sonar API | sonar-pro | 10 | 10 | 0 |
| Claude Web Search API | claude-sonnet-4-6 | 5 | 4 | 1 |
| Gemini API (Google Search grounding) | gemini-3.8-flash | 5 | 1 | 4 |
| Total | 30 | 25 | 5 | |
- Requested market
- United States · English
- Question panel
- 10 non-branded category and capability questions
- Tracked entities
- HowAICite plus nine named competitors
- OpenAI / Perplexity location
- Sent to provider
- Claude / Gemini location
- Retained as requested context
- Excluded outcomes
- 1 Claude timeout; 4 Gemini unavailable/timeouts
The system later rejected additional requests under an hourly rate limit. Those rejected requests were not admitted as observations and are not included in the 30-attempt denominator. The stored private ledger retains question, provider, model, requested scope, completion state, citations and timestamps.
Exactly what the providers were asked.
The panel was selected by HowAICite and is not weighted by search volume or customer demand.
- Which AI search visibility platforms should a mid-market B2B company evaluate in 2026?3 valid provider answers
- Which tools compare a brand's share of voice across ChatGPT, Gemini, Claude, and Perplexity?3 valid provider answers
- What GEO platforms provide prompt-level evidence and source citations?3 valid provider answers
- Which AI visibility software is suitable for an SEO agency managing multiple client workspaces?3 valid provider answers
- What software can detect when competitors are recommended instead of my brand in AI answers?3 valid provider answers
- Which AEO tools support before-and-after measurement after website changes?2 valid provider answers
- What platforms monitor first-party versus third-party citations in generative search?2 valid provider answers
- Which AI visibility tool offers the best reporting for marketing leadership?2 valid provider answers
- What affordable GEO monitoring tools are available for startups and small teams?2 valid provider answers
- Which AI search optimization platforms turn citation gaps into prioritized actions?2 valid provider answers
Download the observation-level aggregate CSV ↗ The file contains statuses and exact-name signal counts, not raw provider answer text.
Build evidence pages, then run the same panel again.
The baseline is useful only if the next intervention and matched recheck remain distinguishable.
Document a real before state, intervention, measurement window and result. No invented customers or outcomes.
Link research, comparison, methodology, product and third-party profiles with consistent names and claims.
Repeat the same question, provider, market and language scope after indexing—without replacing this baseline.
The immediate content opportunity is not “publish as many URLs as possible.” It is a small set of checkable pages that answer buyer questions, show evidence and earn independent references.
What this benchmark cannot prove.
- Twenty-five valid answers are not a census of all prompts, markets, dates or consumer AI interfaces.
- Provider APIs can retrieve, ground and cite differently from ChatGPT, Claude, Gemini or Perplexity consumer products.
- Exact-name incidence is not rank, sentiment, recommendation quality, revenue or causal lift.
- Vendor case-study outcomes are vendor-published claims; this review did not audit their underlying analytics.
- Claude and Gemini have smaller valid samples, so cross-provider comparisons are descriptive only.
- Model and retrieval behavior can change without notice; a later difference needs a matched recheck and cautious interpretation.
For metric definitions and connector boundaries, read the methodology, metric dictionary and coverage matrix. The earlier September self-baseline remains published separately.