AI search platforms review: evidence, fit and limitations
This AI search platforms review compares five vendors using public documentation—not unpublished “real client testing.” Choose LightSite for a website-delivery evaluation, Peec for visibility workflows, Otterly for audits and log analysis, Profound for answer intelligence and content operations, or Scrunch for diagnostics and agent-facing delivery. Validate each fit in your own pilot.
Disclosure and evidence
Sources checked on . This is a public-documentation comparison, not a hands-on benchmark. LightSite publishes this guide and sells one of the products compared. Inclusion is editorial, and the order is not a measured performance ranking.
We compare the job supported, the available evidence and the work left to your team. Product pages establish what vendors advertise; they do not establish reliability, customer outcomes or contract entitlements. No private customer material or competitor test outputs are used. Numeric paid prices are omitted because currency, billing term, usage and regional offers differ.
Otterly’s main feature and pricing pages returned access errors during this review; its accessible help articles support the audit and log-analytics descriptions. Some vendor sites expose features without a clear plan boundary. Treat an unverified cell as a question for procurement, not proof that a feature is absent.
Earlier versions described client testing and subjective onboarding experiences without publishing dates, prompt sets or outputs. Those claims have been removed from this page, its title and its search-engine version. The absence of records here does not prove no testing ever happened; it means this review cannot substantiate that claim.
Five platform trade-offs
| Platform and source | What the documentation supports | What to validate in a trial |
|---|---|---|
| LightSite AI | Connect public business information delivery with crawler visits, AI referrals, answer benchmarking and website improvements. | Requires website integration and an agreed scope of work; not the simplest choice for a reporting-only brief. |
| Peec AI | Track visibility, position, sentiment and citation sources. The current plan page also lists agent analytics, AI referrals and role-based permissions. | Do not assume the reporting plans include every agent feature. Ask which integrations and actions are included. |
| Otterly.AI | Its documentation describes server-log analytics with Netlify, Cloudflare, WordPress, webhook and file-upload connectors. GEO Audit checks crawlability, content and query fan-out. | Logs must be connected. Audit recommendations are not evidence of a deployed fix; confirm who implements each change. |
| Profound | The feature page describes sentiment, citations, competitive benchmarking, server-log analytics and agents that research, write and publish with review checkpoints. | Not merely a dashboard. Validate the required log access, content connector, plan limits and review controls in your own pilot. |
| Scrunch | Audits pages, identifies content gaps and describes serving an agent-facing version at the edge through AXP. | Edge delivery requires a technical rollout. Ask which pages, agents, analytics and rollback controls your contract covers. |
LightSite AI — buyer fit and limitations
Evaluate the connection between infrastructure, intelligence and execution. Public discovery and documentation pages, route-level server HTML and structured-data generation provide examples you can inspect. They do not establish that every customer integration has the same behavior.
For a reporting-only project, Peec or another narrower deployment may reduce integration work. Profound also documents content agents and crawler analytics, while Scrunch documents edge delivery; neither should be dismissed as merely a dashboard.
Sources: Official features / documentation · Current plans; checked 10 October 2026.
Peec AI — buyer fit and limitations
The current plan page goes beyond the old “monitoring only” description: it lists agent analytics, referrals, crawlability auditing, integrations and agent skills as well as visibility reporting. Its SSO documentation specifies SAML and explicitly excludes OIDC and SCIM.
Demonstrate a project report, a restricted user and any proposed agent action. A feature appearing in a broad comparison grid does not establish its availability in your selected plan or that it publishes content for you.
Sources: Official features / documentation · Current plans; checked 10 October 2026.
Otterly.AI — buyer fit and limitations
Accessible help articles describe a GEO audit with crawlability, content and query fan-out, plus actual server-log analytics. The log documentation explains connector choices and distinguishes historical file uploads from live collection. This is concrete evidence against the earlier claim of “no attribution or traffic connection.”
Ask to inspect an uploaded log period and one live connector. Validate identified bots, missing events and the implementation owner for each audit recommendation. Main feature/pricing pages were inaccessible to this review; no price or usability verdict is claimed.
Sources: Official features / documentation · Current plans; checked 10 October 2026.
Profound — buyer fit and limitations
The current feature page describes answer intelligence, sentiment, competitive comparisons, prompt research, log-based agent analytics and content agents with human checkpoints. The enterprise page describes SSO and SOC 2 Type II. This is broader than the earlier enterprise-dashboard characterization.
Run one content draft through a review and publishing connector. Request a log-source demonstration and the security report. We did not measure onboarding speed, product polish, value for money or forecast performance.
Sources: Official features / documentation · Current plans; checked 10 October 2026.
Scrunch — buyer fit and limitations
Current pricing and site-diagnostics pages describe audits, content generation and AXP agent-facing delivery; Enterprise adds brand workspaces, API and SSO. This supports evaluating Scrunch as more than monitoring alone.
Inspect exactly what an agent receives for one approved page and how updates, caching, rollback and visitor separation work. An edge-delivery feature is not proof of a complete cross-site agent journey or successful purchase tracking.
Sources: Official features / documentation · Current plans; checked 10 October 2026.
A reproducible pilot you can run
The following is a proposed protocol, not a test we have completed. Replace the bracketed fields with your real category, country, vendors and requirements; do not use a vendor’s preferred prompt list as the only sample.
| Intent | Prompt template | What to record |
|---|---|---|
| Category discovery | Which [category] providers serve [country] for [use case]? | Brand mentions, recommendation wording and cited URLs. |
| Comparison | Compare [your brand] and [competitor] for [requirement]. What are their limitations? | Factual errors, caveats, sources and whether each claim can be verified. |
| Purchase research | Which [category] vendors support [integration] and [security requirement]? Link to official evidence. | Official versus third-party sources and missing or outdated claims. |
| Action path | On [your public page], find the approved way to request [demo or quote]. | Pages read, CTA discovery and blockers. Stop before submission unless explicitly authorized. |
Run each question three times per selected assistant over a defined week using fresh conversations, one country/language and saved model/interface settings. This is a manageable pilot design, not a statistically representative sample of all searches. Record the date, raw answer, URLs and collection failures for every run.
- Score mentions and citations separately; count unsupported product claims and incorrect business facts.
- Inspect the cited pages rather than assuming a source URL supports the answer.
- Compare the report export with the saved raw answers. Document disagreement and missing responses.
- Test a single approved page update and repeat the same collection protocol; changes in answers alone do not establish causation.
- Keep observed crawler requests, human referrals, agent interactions and confirmed transactions as separate measurements.
Publish results only when the raw evidence and permission to share it exist. Until then, use the protocol to evaluate fit, not to claim that our product or any competitor has won a benchmark.
Read the metrics without overstating them
An answer-monitoring system can reliably store what it sampled. The limitation is representativeness: users may see different answers because of context, wording, location, model or timing. That does not make all mention tracking meaningless; it makes consistent sampling and uncertainty reporting necessary.
Server logs establish observed requests, not that a crawler used every fact or caused a recommendation. Referral analytics establishes visits your setup can recognize, not every assistant-influenced sale. Agent-action data is useful for finding blocked task paths, but a form attempt is not a confirmed lead and a checkout visit is not a purchase.
For a deeper discussion, read the approaches to tracking mentions. For a real task-path use case, see the hotel booking journey article.
Frequently asked questions
Was this review based on real client testing?
No reproducible competitor test dataset, dated prompts or sample outputs are available in the repository for this review. The current comparison is based on linked public documentation, not a claimed hands-on performance benchmark.
Can AI mention tracking be accurate?
It can accurately record the answers obtained in a defined sample. It does not establish what every user sees; wording, country, model, time and repeated runs must be documented.
How do I compare two platforms fairly?
Use the same prompts, countries, models and collection window; export raw answers and citations, then test one workflow with the same page and approval conditions. Record missing data rather than substituting a zero.
Does crawler analytics prove a conversion?
No. A crawler request proves a request was observed, and a contact click proves an interaction was recorded. Confirmed sales or bookings need the relevant transaction or CRM record.
Vendor sources
Official documentation checked on 10 October 2026; these references support the comparison, not an endorsement or a performance ranking.
Related buying guides
Best GEO tools by task and enterprise procurement criteria address different buying questions from this five-platform review.