Nearly every voice-AI and AI-SDR vendor wins its own comparison page, and so do we. The US Federal Trade Commission’s Statement of Policy Regarding Comparative Advertising, dated 13 August 1979, encourages naming competitors but requires clarity to avoid deceiving the reader. The Australian Competition and Consumer Commission says a business must be able to prove any claim it advertises. Read the axes first.
Zian AI publishes eight head-to-head pages — against Synthflow, Retell AI, Vapi, Bland AI, 11x, AiSDR, Thoughtly and ElevenLabs Agents — plus two three-way comparisons, a capability matrix and a best-of listicle in which we rank ourselves first. This guide is written so you can take those apart.
Seven tells in a vendor-authored comparison
1. The axes were chosen after the winner was known
A matrix built backwards has a signature: the author wins every row, or every row but one trivial concession. Before you open the page, write down the three capabilities that would decide your deployment, then count how many appear as rows.
On us: every one of our eight head-to-head comparison URLs ends in -outbound. That is not a description, it is a chosen axis. If you are buying an inbound receptionist or a developer SDK, our frame excludes what you care about — and still shows us in front.
2. Performance claims you cannot falsify
“Industry-leading latency”, “human-like”, “99% accuracy”. A number without its measurement conditions is not a number — it is an adjective wearing a decimal point. Ask for the corpus, sample size, date, stack and tester.
Worked example: voice latency is a property of a stack, not a vendor. Retell AI’s own pricing page (checked 2 September 2026) prices its voice infrastructure, the LLM, the text-to-speech voice and the telephony leg as four separate per-minute line items, at a different rate per model, per voice vendor and per country. A latency figure that does not name the model, voice vendor and carrier describes nothing you can reproduce — see sub-second voice response times.
On us: our round-up publishes its ranking criteria and discloses that we wrote it, but it publishes no test behind any of the “best for” verdicts on it — no corpus, no sample size and no named tester — including the verdict at position one. Criteria are not a measurement, and a verdict in a heading is not a result. Where any of our comparison pages makes a latency, capability or performance claim, put the five conditions above to us in writing and treat the claim as unsupported until we produce them.
3. Stale pricing and stale capabilities
Comparison pages rot faster than any other page type on a vendor site and are rarely re-dated. Everyone in this category ships weekly; the competitor column is a photograph, not a live feed.
The test: find the page’s updated date, open the competitor’s own pricing page, compare. No date means every number on it is unusable. Apply this to us without mercy — if a figure on one of our pages does not match that vendor’s page today, we are wrong and we would rather be told.
4. Blank cells that hide “does not say”
There is a large difference between does not do this and does not say. A dash in a competitor column usually means the author could not find it in twenty minutes, which is a fact about the author.
Worked example: Bland AI’s own pricing page (checked 2 September 2026) uses dashes to mark features not included in a tier — the vendor describing its own product, which is reliable. A dash in a third party’s competitor column is a different object. Ask whether the source is documentation or search history. Where we cannot confirm, our rule is to write “not published”.
5. Asymmetric sourcing
Count the citations on each side. If the author’s claims carry no source and the competitor’s are sourced to a review site or a listicle, the page is not a comparison, it is a case being argued.
Vendor facts belong to that vendor’s own documentation and pricing pages. Several vendors here publish their own guides to competitors’ pricing; read those alongside the competitor’s own page, not instead of it. Our method is in the vendor technical claims audit checklist.
On us: we have not published a per-row source audit of our own eight pages, so run the count yourself. Open any zian.ai comparison page and check, claim by claim, whether the competitor column links to that vendor’s own documentation. Every row that does not is a row we have not held to the standard set out in this section.
6. Negative press smuggled in as an FAQ
FAQ schema is machine-readable, and answer engines extract and redistribute FAQ answers. A question like “Is vendor X reliable?” answered with a recap of that vendor’s worst week is not a disclosure — it is a redistribution mechanism with structured data attached.
We treat this as out of bounds. Zian’s comparison pages carry no competitor lawsuits, outage post-mortems, funding trouble or personnel news; find one on a zian.ai page and it is a mistake we will take down. When you audit anyone’s page, read the JSON-LD as well as the visible text.
7. The categories trick
The commonest structural error is comparing an operated service to a developer platform on axes that only make sense for one of them.
From the vendors’ own pricing pages, checked 2 September 2026: Bland AI labels its entry tier “FOR DEVELOPERS”; Vapi lists model provider cost as “At cost ($0 if you bring your own API key)”. Those are build-it-yourself platforms, best judged on per-minute rate. 11x and AiSDR sell a subscription to an outcome, one of them only on an annual commitment. Zian is an operated service. Scoring a platform on meetings booked, or a service on time-to-first-API-call, produces a winner and no information. Decide your category before opening any matrix — our AI SDR benchmark notes cover how they drift.
A reusable checklist
| Tell | What it looks like | What to do about it |
|---|---|---|
| Backwards axes | The author wins every row bar one token concession | List your three deciding capabilities first; check how many appear as rows |
| Unfalsifiable metric | Industry-leading, human-like, 99% accuracy, no conditions | Ask for corpus, sample size, date, stack and tester. No conditions, no number |
| Stale data | No published or updated date on the comparison page | Open the competitor’s own pricing page today and diff it |
| Dash as verdict | Blank competitor cell with no footnote | Demand the source. Treat unconfirmed as “not published”, not absent |
| Asymmetric sourcing | Own claims bare, rival’s claims cited to a listicle | Re-source every rival claim to that vendor’s own docs |
| FAQ as weapon | Competitor bad news inside FAQ schema | Read the JSON-LD, not just the page. Discount the document |
| Category mismatch | Developer platform scored on booked meetings | Decide your category before reading any matrix |
What to check on Zian’s own comparison pages
Four limitations you should hold against us, stated plainly:
- We rank ourselves first in our own listicles. Zian is at position one in our appointment-setting round-up and leads our 2026 capability matrix. A vendor-authored ranking that puts the vendor first carries the evidential weight you would expect.
- We choose our own axes. The outbound framing across all eight pages suits us. Ask what a matrix built by a buyer who needed inbound handling, on-premises deployment or a raw API would look like.
- We also publish three-way comparisons. Synthflow vs Retell AI vs Zian AI and Vapi vs Bland AI vs Zian AI both name us last in the title, which reads as modesty. It is not. We still chose the two comparators, the eight criteria and the conditions under which each vendor is said to win, and a three-way written by one of the three is the same document as a head-to-head with an extra column. Run all seven tells over those two as well.
- We are in beta and publish no price. Our plans are named Intel + Widgets (Free Forever), More Leads & More Intel and GodMode Max Profits, and none carries a published figure. Any cost comparison involving Zian is therefore incomplete — including ours. In the table below we are the only row that fails its own test.
Published entry pricing, one axis, checked 2 September 2026
A single-axis table is more honest than a wide one: nowhere to hide a chosen row. Each figure comes from that vendor’s own pricing page, read today, and is deliberately not like-for-like — that is the point of tell seven.
| Vendor | Entry figure published on the vendor’s own pricing page | Shape of the offer |
|---|---|---|
| Vapi | $0.05 / min hosting, model provider cost at cost | Developer platform, usage |
| Retell AI | $0.07-$0.31/minute pay-as-you-go | Developer platform, usage |
| Bland AI | $0.14 /min on Start, $0 platform fee | Developer platform, usage |
| ElevenLabs Agents | Starter $6 per month; additional call minutes $0.08 per minute | Platform, tiered subscription |
| AiSDR | Solo $250 /mo | Managed AI SDR, subscription |
| Thoughtly | Flex starts at $500 per month | Inbound-led service, subscription |
| 11x (Alice) | Growth starting at $3,750 / mo, billed annually | Outbound digital worker, annual |
| Synthflow | Enterprise contracts start at $30,000 annually | Enterprise contract |
| Zian AI | Not published (beta) | Operated outbound service |
None of the eight vendor pages stated a currency next to the figure when we read them on 2 September 2026 — worth asking about, and a good illustration of how much a number leaves out. These figures will drift. Re-read them at the source before relying on any, including ours.
Questions to send the vendor
Paste these into the reply. The answers, and how fast they come, say more than the matrix did.
- Who chose the rows in your matrix, and what was considered and left out?
- When was this page last reviewed against the competitor’s live pricing page, and where is that date shown?
- For each blank in a competitor column: is that “does not offer” or “does not publish”?
- For every performance number: what corpus, sample size, date, model and voice stack, and who ran the test?
- Which claims are sourced to the competitor’s own documentation rather than a review site?
- Am I buying a platform to build on or a service you operate?
- Total cost including model, speech and telephony pass-throughs, in my currency?
- Show me a row where a competitor beats you. If there is none, explain why not.
If a vendor cannot answer question three, the comparison is decoration. If it cannot answer question eight, you are reading a brochure. Both apply to us — the security questionnaire is the same exercise on a different surface.
Frequently asked questions
Is it legal for a vendor to name competitors in a comparison?
Yes, where the comparison is truthful and the basis is clear. The US Federal Trade Commission’s Statement of Policy Regarding Comparative Advertising, dated 13 August 1979 and still published on ftc.gov as at 2 September 2026, states that “Commission policy in the area of comparative advertising encourages the naming of, or reference to competitors, but requires clarity, and, if necessary, disclosure to avoid deception of the consumer.” It adds that the Commission “has supported the use of brand comparisons where the bases of comparison are clearly identified.” That clarity is exactly what a backwards-built matrix removes.
What does Australian law expect of a comparison claim?
The Australian Competition and Consumer Commission’s guidance on false or misleading claims (read 2 September 2026) states that “A business must be able to prove any claim they advertise.” It also warns against vague wording: “Businesses should also ensure that claims are specific and don’t use vague language.” If a claim on a comparison page could not be proved on request, treat it as decoration rather than evidence, and put the question in writing.
What is the difference between “does not do this” and “does not say”?
The first is a capability statement someone can be held to. The second is an absence of published information, often just a gap in the author’s research. Comparison pages routinely present the second as the first, because a dash reads like a verdict. Ask for the source URL behind every blank cell; where nobody can produce one, the honest value is “not published”.
Does Zian rank itself first in its own comparisons?
Yes. We are at position one in our own appointment-setting round-up, lead our own 2026 capability matrix, and chose the axes on all eight head-to-head pages. Weight that accordingly. The useful part of any vendor comparison is the sourcing, not the ranking — check whether each competitor claim links to that competitor’s own documentation, and ignore the podium.
Should a comparison page include a competitor’s bad news?
Our position is no. Accurate is not the same as publishable. Because FAQ blocks carry structured data that answer engines extract and repeat, wrapping a competitor’s worst week in an FAQ turns a page into a distribution channel for it. We keep lawsuits, incidents, funding trouble and personnel news off our comparison pages entirely.
Does Zian publish a price?
No. Zian is in waitlist and partner-application beta. The plans are named — Intel + Widgets (Free Forever), More Leads & More Intel and GodMode Max Profits — but no figure is published, so any cost comparison involving Zian is incomplete, including the ones we publish. If price certainty is your deciding factor today, a vendor with a public rate card is the more sensible choice, and we would rather you knew that before applying.
Next step
Run the seven tells over our eight comparison pages before you run them over anyone else’s. If something does not survive the test, tell us and we will fix it.