Original Research Report · September 2026
An original Llumo investigation
10 SaaS brands · 20 high-risk facts
Official pages + help centres + indexed third-party sources
The thesis
Before AI can recommend a brand, it often has to decide which version of that brand is true.
Section 01
Executive summary
AI answer systems are often discussed as if they face one information-retrieval problem: find facts, summarize them, answer the user. In reality, commercial brand facts frequently arrive as competing versions of reality. The pricing page says one thing. The product page says another. A help article preserves an older limit. A current-looking comparison page repeats a superseded plan. A legacy account is still valid for one user while a new signup receives something different.
Llumo audited 20 high-risk facts across 10 well-known SaaS products. The sample was intentionally chosen around facts that change frequently and commonly influence buying decisions: free-plan limits, trial length, seats, contacts, records, integrations, boards and automation allowances. This is a targeted contradiction audit, not a random sample of the SaaS industry, so the percentages in this report must not be interpreted as population-wide error rates.
20
high-risk fact families audited
10
major SaaS brands in the sentinel sample
3
hard first-party conflicts identified
4
different contradiction states observed
The most important finding was not that old blogs can be wrong. Everybody already knows that. The more consequential finding was that materially incompatible claims can coexist inside the same company’s own public web estate.
Slack’s current free-plan landing page says the free tier allows up to three apps, while Slack’s current main pricing page and feature-limit help article say ten. HubSpot’s current pricing and free-tools pages describe the free account as supporting up to two users and 1,000 contacts, while other current HubSpot product pages still advertise unlimited users and up to one million contacts. These are not obscure scraped copies. They are first-party pages on the brands’ own domains.
The web does not contain one canonical version of a company. It contains a stack of claims with different dates, contexts, authorities and update speeds.
The audit also found a second layer: stale but still accessible information. Mailchimp reduced its free tier to 250 contacts and 500 monthly sends in February 2026, yet pages published or updated in 2026 continued to state the previous 500-contact / 1,000-send allowance. Zapier’s current pricing now advertises unlimited Zap workflows on the free tier, while a March 2026 pricing guide and a September 2026 automation consultancy page still state a five-Zap limit. Airtable’s current free plan allows 1,000 records per base, but older high-intent community and instructional pages remain indexed with the previous 1,200-record limit.
A third layer looks like contradiction but is actually context. Asana’s current Personal plan allows two users for new signups, while legacy Personal users may retain ten. Semrush’s normal SEO trial is seven days, but some partner or add-on contexts legitimately offer fourteen days. An AI system that collapses those conditions into a single answer can turn two individually true statements into one misleading statement.
Observed status across the 20 targeted fact families
Hard 15%
Stale 25%
Context 10%
Clear 50%
These proportions describe this deliberately high-risk sentinel sample only. They are not estimates of error prevalence across SaaS or the web.
The implication for AI visibility is fundamental. Before an answer engine can decide whether a brand is good, cheap, free, suitable or worth recommending, it may first need to arbitrate among competing statements about what the brand even is today. That upstream disagreement is what this report calls The Contradiction Layer.
Section 02
The hidden data problem behind AI answers
Most AEO and GEO strategy starts downstream. It asks which sources AI cites, whether the brand appears, how often it is recommended, or what wording accompanies the recommendation. Those are useful questions. But they assume the source environment itself is coherent.
For many brands, it is not.
Software products are particularly exposed because commercial facts change quickly. Free plans shrink. AI features move between tiers. seat allowances change. Pricing is regional. legacy customers remain on older entitlements. promotional trials appear through partners. Help centres and product pages are managed by different teams. Affiliate content can stay live long after a change. Community answers can rank for years.
Traditional search makes that mess visible. A buyer may open three tabs and notice that the numbers differ. An AI answer is more compressive. It normally returns one fluent synthesis. That creates a risk: contradiction can disappear from the interface even when it still exists in the evidence underneath.
Contradiction is not the same as hallucination. An AI system can produce a false or misleading answer without inventing anything. It may faithfully repeat a real page that is stale, apply a legacy entitlement to a new customer, or choose one of two first-party pages that disagree.
Why “just use the official website” is not enough
The obvious answer is to privilege first-party sources. The audit shows why that rule is incomplete. When first-party pages disagree, “official” identifies ownership but not necessarily freshness or applicability. A pricing page may be current while a long-lived product landing page preserves copy written under a previous commercial model. Both are official.
Authority therefore has to be decomposed. A useful answer system needs to care not only about who published a claim, but also:
Recency
When was the claim last materially verified?
Specificity
Does it describe the exact plan, geography, account type and product?
Canonicality
Is the page intended to define the fact, or merely mention it?
Version
Does the statement apply to current, legacy or promotional users?
Observed product state
Does the live product behave the way the documentation says?
Corroboration
Do other current authoritative sources agree?
Company changes product
→
Official pages update unevenly
→
Third parties lag
→
Search index keeps old pages
→
AI must arbitrate
Section 03
Methodology
This is a sentinel audit designed to expose contradiction patterns, not to estimate how often every SaaS fact on the internet is wrong.
Sample
We selected ten widely used SaaS products whose plans, limits or commercial structures have changed over time: Slack, HubSpot, Asana, Zapier, Mailchimp, Airtable, Miro, Semrush, Trello and monday.com.
For each product we selected two high-risk facts that are both commercially meaningful and likely to be repeated across the web. That produced 20 fact families.
10 brands × 2 high-risk facts = 20 audited fact families
Source hierarchy
Classification
Every fact family received one of four statuses:
Two current authoritative statements are materially incompatible under the same apparent conditions.
A superseded claim remains publicly accessible or is repeated on a page that can still influence discovery.
Different claims can both be true, but only for different user cohorts, offers or product contexts.
The checked sources were materially aligned or any differences were straightforwardly explained.
What counts as a contradiction?
We excluded trivial wording differences. “Three editable boards” and “three active boards” are not automatically treated as different facts. We also excluded monthly-versus-annual price differences where billing cadence was clearly labelled.
We included a contradiction only where two claims would lead a reasonable buyer to materially different expectations: two users versus unlimited users; 250 contacts versus 500; five workflows versus unlimited workflows; three integrations versus ten.
Selection bias is intentional. We deliberately chose volatile facts and products with long-lived documentation ecosystems. The result is a map of contradiction failure modes. It is not a claim that 50% of SaaS facts on the internet are contradictory.
1
Current canonical first-party pages.
Pricing pages, current help-centre documentation and plan-comparison pages were treated as the strongest baseline where they agreed.
2
Other first-party pages.
Product landing pages, older help content, community answers, announcements and historical first-party material were checked for incompatible statements.
3
Indexed third-party pages.
Current and older independent pricing guides, reviews and specialist articles were checked where they surfaced materially different claims.
4
Context resolution.
We attempted to determine whether apparent conflict could be explained by legacy eligibility, promotional access, annual versus monthly billing, partner links, product variants or phased rollout.
Section 04
A taxonomy for the Contradiction Layer
1. First-party split
The strongest form occurs when the company’s own live pages disagree. This is the hardest case for simple “trust the official source” logic because ownership no longer resolves truth.
2. Stale echo
A fact was once correct, changed, and continues to echo through indexed pages. Stale echoes are especially dangerous because they often look credible: they may be dated recently, rank well, or sit on a trusted domain.
3. Legacy bleed
A historical entitlement remains valid for one cohort and leaks into general descriptions of the current product. “Free supports ten users” may be true for an account created before a cutoff and false for someone signing up today.
4. Offer collision
The same product can legitimately have different trial lengths or prices through standard checkout, a partner campaign, a regional promotion or a product-specific add-on. An answer that strips away the conditions converts context into contradiction.
5. Documentation/product drift
Documentation describes one behavior while a live user encounters another. This may be a bug, rollout, experiment or delayed docs update. From an answer engine’s perspective, however, it creates unresolved evidence.
The Brand Contradiction Risk model
For practical monitoring, Llumo proposes treating contradiction risk as the combination of four factors:
Brand Contradiction Risk ≈ Source disagreement × fact importance × source visibility × time-to-correction
A wrong limit buried in a ten-year-old forum with no search visibility may carry little practical risk. A wrong free-plan claim on a current high-ranking product page carries much more. The severity is not simply whether an inconsistency exists; it is whether buyers and AI systems are likely to encounter it.
Section 05
The 20-fact audit
The table below records the status assigned to each audited fact family. “Clear” does not mean every page on the internet agrees; it means the pages checked for this pilot did not produce a material unresolved conflict.
Brand
Fact family
Current baseline
Status
Reason
Slack
Free-plan app integrations
10 apps on main pricing/help pages
Hard conflict
Separate current Slack free-plan page says 3 apps.
Slack
Searchable message history
90 days
Clear
Current plan pages checked align.
HubSpot
Free users
Up to 2 on pricing/free-tools pages
Hard conflict
Current HubSpot CRM pages advertise unlimited users.
HubSpot
Free contacts
1,000 contacts on current free-tools FAQ
Hard conflict
Current HubSpot product pages advertise up to 1 million contacts.
Asana
Personal-plan seats
2 for current signups; 10 for eligible legacy users
Stale / versioned
Older official/community material still states 10 or 15 without current-signup context.
Asana
Starter entry price
$10.99/user/month starting point
Clear
No material unresolved conflict in checked current sources.
Zapier
Free-plan Zap count
Unlimited Zap workflows on current pricing
Stale conflict
2026 third-party pages still state 5 Zaps.
Zapier
Free monthly task allowance
100 tasks
Clear
Current sources checked align on 100.
Mailchimp
Free contacts
250
Stale conflict
2026 pages remained indexed with the previous 500-contact limit.
Mailchimp
Free monthly sends
500
Stale conflict
2026 pages remained indexed with the previous 1,000-send limit.
Airtable
Free records per base
1,000
Stale conflict
Older community/instructional pages still state 1,200.
Airtable
Free API calls
1,000/workspace/month
Clear
Current official overview is explicit.
Miro
Free board creation/editability
3 editable boards; pricing also describes unlimited creation with only 3 editable
Context / drift
September 2026 user report shows live UI blocking a fourth board despite docs wording.
Miro
Free team-member count
Unlimited / 1+
Clear
Current pricing/help pages materially align.
Semrush
Trial duration
7 days for standard SEO + AI Search plans
Contextual
14-day offers exist for some partner/add-on contexts.
Semrush
Standard SEO trial availability
Available, subject to eligibility
Clear
Current official trial and subscription pages align.
Trello
Free boards
10 open boards/workspace
Clear
Current official pricing/support pages align.
Trello
Free collaborators
10/workspace
Clear
Change was phased in during 2024 and current documentation is clear.
monday.com
Free seats
2
Clear
Current support documentation is explicit.
monday.com
Free base item allowance
200, expandable by referrals
Clear
Potential “up to 1,000” interpretation is explained by the referral mechanism.
Three of the 20 fact families produced hard same-brand contradictions, five produced stale conflicts, two required meaningful context to reconcile, and ten were materially clear in the sources checked.
Again, the numerical split is descriptive of the targeted sample. The value of the dataset is the pattern: contradiction can originate at every layer of the public information stack, including the layer most marketers assume is safest.
There may be an internal rollout or an outdated landing page behind the difference. The public pages themselves do not resolve it for a buyer landing on them independently. From the outside, both are authoritative.
This is exactly the sort of situation where an AI answer can appear confidently wrong while relying on a genuine official source. If it says “Slack Free supports three integrations,” it can cite Slack. If it says ten, it can also cite Slack.
This is the core Contradiction Layer problem.
Domain authority cannot resolve two incompatible claims when both claims live on the same authoritative domain.
The difference is not a minor plan-description nuance. A two-person startup and a twenty-person sales team would make different buying decisions based on those claims. Likewise, 1,000 contacts and one million contacts describe fundamentally different free products.
The contradiction has already escaped the HubSpot domain. Several 2026 third-party CRM guides still repeat “unlimited users” as a current HubSpot advantage, while other current reviewers explicitly warn that the free tier is now capped at two users and 1,000 contacts.
For an AI system, corroboration can therefore make the problem worse rather than better. If multiple external articles copy the stale first-party claim, a naïve “more sources agree” rule can reinforce the older reality.
Corroboration is only valuable when the sources are independently current. Ten copies of the same stale fact are not ten independent confirmations.
A March 2026 independent pricing page still described “Free” as 500 contacts and 1,000 sends. A March 2026 EmailCloud article likewise called 500 contacts and 1,000 sends the current plan despite the February change. Other publishers updated later and now explicitly document the cut.
This shows why timestamps are necessary but insufficient. A page labelled “2026” is not necessarily current for a fact that changed during 2026. Even a recently published article can be temporally wrong if its source data was collected before the product change or copied from an older comparison.
The correction problem
Stale facts do not disappear when the vendor updates its pricing page. They remain in:
search-indexed comparison pages
affiliate tables and cached snippets
community answers and forum replies
screenshots and PDFs
articles whose “updated” label does not guarantee every fact was rechecked
AI-generated summaries that may themselves become new sources
The correction event is therefore not a single update. It is a propagation process.
Yet independent pages published in 2026 still describe the free tier as limited to five Zaps. One pricing guide updated in March 2026 lists “5 Zaps.” A UK automation consultancy page published on September 1, 2026—only weeks before this report—also tells beginners that the free plan includes “up to five Zaps.”
This is the sort of contradiction most likely to enter AI answers because the stale pages are not obviously ancient. They use the current year, discuss current products and contain otherwise plausible details. Fresh-looking packaging can conceal old product assumptions.
That already creates two legitimate answers to “How many people can use Asana for free?” depending on who is asking.
The web adds a third historical layer. Older Asana content still indexed in 2026 describes free Organizations with teams of up to 15 members. A 2024 community answer says the Personal plan supports ten users. A 2024 government procurement pricing PDF likewise records Personal as “up to 10 users.” Those statements accurately reflect earlier product states but are not the correct default for a new 2026 signup.
Older free model: 15
→
Legacy Personal: 10
→
Current signup: 2
This is a temporal/versioning problem rather than a simple error. A strong answer needs to preserve the condition:
Correct synthesis
New Asana Personal accounts support two users; some qualifying legacy Personal accounts retain higher limits.
Flattening the history into one number is where the misinformation begins.
The previous 1,200-record limit has a much longer web history. Airtable community pages from 2016 through 2021 repeatedly state 1,200 records per free base. Independent instructional material still accessible online repeats the same number. Some of these pages rank because they directly answer high-intent questions such as “Is there a limit to the number of records?”
This matters because question-answer pages are structurally attractive retrieval candidates. They match natural-language queries almost perfectly. A ten-year-old answer can therefore remain semantically relevant while being factually obsolete.
Semantic relevance and temporal correctness are different dimensions. Search systems—and AI systems built on retrieval—need both.
In September 2026, however, a Miro community user reported that the live product blocked creation of a fourth board and displayed a message saying the Free plan includes three boards and that an upgrade is required to create unlimited boards.
One community report cannot establish the universal product rule. It could reflect a rollout, experiment, account-specific state or bug. That uncertainty is precisely the point: current documentation and observed product behaviour can temporarily diverge.
This category should not be labelled “wrong” without further testing. It should be labelled unresolved and monitored.
But fourteen-day access also exists in legitimate contexts. Semrush documents a fourteen-day Agency Partners Platform add-on trial, and historical affiliate/partner materials describe extended fourteen-day trials available for promotion.
There is no inherent contradiction if the context is preserved. The failure occurs when a comparison page or AI answer collapses a partner-specific or add-on-specific offer into a universal statement such as “Semrush has a 14-day free trial.”
Context is data.
Trial length is not only a number. It is a tuple: product × user eligibility × acquisition channel × date × region.
Section 14
Why the Contradiction Layer matters for AI
An AI answer engine does not simply retrieve facts. In many commercial questions it performs an implicit truth-arbitration task.
1. Fluent answers hide evidence disagreement
Search results expose disagreement visually: users see multiple links, dates and snippets. Conversational answers often collapse that disagreement into one sentence. The interface therefore makes the answer easier to consume while making the underlying conflict harder to notice.
2. Citation can create false reassurance
A cited answer can still be misleading if the cited page is stale or context-specific. The Slack and HubSpot cases go further: a source can be first-party and still conflict with another first-party source.
3. Repetition can amplify stale truth
When one obsolete claim is copied into many comparison pages, source count becomes a poor proxy for truth. Ten pages can be descendants of one old pricing table. This creates an evidence-cloning problem.
4. Temporal qualifiers are easy to drop
“Existing accounts before November 12, 2025” is crucial context. So is “partner trial,” “annual billing,” “for verified site owners” or “for Business plan only.” Summarization pressure encourages short answers, and short answers are exactly where qualifiers disappear.
5. AI-generated content can feed the next generation of contradiction
If an AI system repeats a stale limit into a blog post, comparison page or knowledge base, the output can become a new indexable source. That creates the possibility of a feedback loop:
This is one reason brand truth management should become part of AEO. Visibility without factual consistency can scale the wrong version of a company.
Section 15
The Brand Truth Stack
To make the problem actionable, Llumo proposes a hierarchy for auditing factual brand claims.
Pricing, plan comparison, checkout, current terms, official product availability.
Help centre, current limits, eligibility, downgrade rules, plan-version notes.
Product pages, solution pages, campaign landing pages, legacy SEO pages.
Reviews, comparisons, directories, press, specialist guides and communities.
The ordering is not absolute. A help page can lag a product rollout. A checkout page can be personalized. A third-party reviewer may test the product more carefully than a marketing page. The purpose is to force explicit arbitration rather than treating all URLs as equivalent.
Claim-level metadata companies should publish
For volatile facts, companies should make the following machine-readable or easily extractable wherever possible:
Effective date
When did this limit or price become valid?
Eligibility
New users, legacy accounts, nonprofits, partners, regions?
Billing basis
Monthly, annual, per seat, per workspace or usage?
Product scope
Which exact product, add-on or bundle does the claim describe?
Supersedes
Which earlier public fact is no longer current?
Canonical source
Which page should external systems treat as authoritative?
Section 16
What companies should do about it
1. Audit facts, not pages
Most website audits work page by page. Contradiction audits should work claim by claim. Pick the 20 facts most likely to affect purchase decisions and search every place those facts appear.
2. Start with volatile facts
Prioritize pricing, free-plan limits, trial periods, features, integrations, locations, support levels, product names, availability, leadership, compliance status and acquisition/rebrand information.
3. Search your own domain as if you were a stranger
Use queries such as:
site:yourdomain.com “free plan” site:yourdomain.com “up to” users site:yourdomain.com pricing site:yourdomain.com trial site:yourdomain.com old-product-name
The HubSpot and Slack examples show why this matters. Teams often know which page is canonical internally. Search engines and answer systems may not.
4. Redirect or annotate obsolete first-party pages
Old product articles can continue to rank because they accumulated links and relevance. If the page is historically useful, add a prominent current-status note and a canonical route to the current plan definition. If it is not useful, consolidate it.
5. Publish change logs for commercial facts
A clear record such as “From 17 February 2026 the free plan changes from 500 contacts to 250” gives search engines, writers and AI systems a better temporal signal than silently overwriting the number.
6. Correct high-visibility third parties
Do not try to purge every old mention on the internet. Rank third-party pages by their likelihood of influencing customers or AI retrieval: search visibility, backlinks, citations, category relevance and recency. Correct the most influential first.
7. Track contradiction as an AEO metric
Llumo proposes monitoring a set of canonical brand facts and testing whether major AI engines reproduce them accurately over time.
Fact Accuracy Rate = correct AI answers ÷ tested factual prompts Contradiction Exposure = high-visibility conflicting sources ÷ audited sources Correction Lag = days between canonical fact change and external-source correction
8. Watch competitor facts too
Recommendation answers are comparative. If an AI believes a competitor still offers a free plan that was removed six months ago, your brand can lose on a false comparison even if your own information is perfectly accurate.
Research Deep Dive
The contradiction lifecycle
Most stale brand facts are not born false. They become false after the company changes and the surrounding information system fails to change with it.
The audit suggests a recurring lifecycle:
Stage 1 · Publication
A company introduces a pricing limit, free tier, trial period or product feature. The fact spreads outward from the official site into reviews, comparison tables, community answers, partner pages and screenshots. At this stage repetition is helpful. It makes the brand easier to understand.
Stage 2 · Change
The company changes the underlying fact. The canonical pricing page is usually updated first because it directly affects conversion and billing. Other pages may belong to different teams, templates or content systems and can remain untouched.
Stage 3 · Asymmetric correction
Some publishers update quickly. Others do not. A page may continue to rank precisely because it is old and authoritative. The source with the strongest search history can therefore be the source least synchronized with the current product.
Stage 4 · Evidence cloning
New content may copy the stale value from existing comparison pages instead of re-verifying the canonical source. Generative writing tools can accelerate this. The stale claim gains a recent publication date even though the underlying fact is old.
Stage 5 · Answer arbitration
An AI system receives a mixed evidence set. If it chooses the current page, the answer is correct. If it chooses the stale page, the answer is wrong. If it notices the disagreement, it may hedge. If it collapses legacy and current plans, the answer can be technically sourced but practically misleading.
Stage 6 · Self-reinforcement
The AI answer itself may be copied into a new article, social post, support answer or comparison page. That creates another retrievable instance of the stale fact and can extend its life.
The contradiction problem is therefore a propagation problem. Correcting the official website is necessary, but it does not instantly correct the public evidence graph surrounding the brand.
What Llumo could measure
A future Contradiction Monitor could maintain a small set of canonical facts for each tracked brand, discover high-visibility sources making claims about those facts, and continuously compare what the web and major AI systems say against the canonical state.
Canonical Fact Set
The current values the company says are true.
Conflict Surface
Pages and domains presenting incompatible values.
AI Accuracy
Which engines currently reproduce the canonical value.
Correction Lag
How long external evidence takes to converge after a change.
Source Influence
Which conflicting sources repeatedly appear in citations or retrieval.
Competitor Distortion
Where a rival is being compared using obsolete pricing or features.
This reframes AI visibility from a pure share-of-voice problem into a truth-maintenance problem. A brand can have high visibility and still lose if the version being amplified is obsolete.
Section 18
The next research phase
This pilot reveals the failure modes. A larger Llumo study could quantify their prevalence and directly test how different answer engines arbitrate them.
Proposed 100-brand contradiction census
100 SaaS brands × 10 canonical commercial facts = 1,000 fact families Each checked against: • canonical first-party page • second first-party source • top independent comparison source • top review/directory source • AI answer across 5 engines
That would create up to 5,000 source-level fact comparisons before the AI-answer layer is even added.
Questions the larger study could answer
Which facts become stale fastest: pricing, seats, integrations or trials?
How often do first-party pages disagree with each other?
Which source type corrects fastest after a product change?
Do AI engines favor canonical pricing pages over old product pages?
Does citation presence improve factual accuracy?
How long after an official change does an AI answer converge on the new fact?
A controlled correction experiment
The strongest follow-up would observe real product changes from day zero. As companies change a plan or feature, Llumo could snapshot the official source, monitor high-ranking third parties and query major AI systems daily or weekly. That would allow a genuine Web Correction Half-Life to be calculated rather than inferred from static pages.
It would also let us test whether the web corrects in layers: official pricing first, help centres next, specialist media next, general comparison sites later, AI answers last—or whether the order is different.
Section 19
Limitations
This report is exploratory and deliberately targeted. The brands and facts were selected because they are commercially important and likely to change. That makes the study useful for discovering contradiction patterns but unsuitable for estimating an industry-wide “error rate.”
The audit is based on publicly accessible web pages retrieved in September 2026. Product experiences may differ by region, account age, experiment, logged-in state, enterprise contract or phased rollout. Where such context was plausible, the report classifies the case as contextual rather than declaring one side false.
The Miro case is based partly on a recent community report of live product behavior. It should be treated as an unresolved documentation/product-drift signal, not proof that every free account behaves the same way.
Old community posts and historical pages are intentionally included because they remain accessible and can be retrieved. Their presence does not mean the company currently endorses the old fact.
This report does not claim that any specific AI model currently retrieves or relies on any particular page cited here. The research examines the public evidence environment that answer systems may encounter. A separate controlled model study would be required to attribute an AI answer to a specific source.
Finally, commercial plans change frequently. Readers should use the linked canonical pages for current purchase decisions rather than treat this report as a permanent pricing reference.
Section 20
Sources
All sources were checked during the September 2026 research window. The list emphasizes pages directly supporting the case studies rather than every page reviewed.
1 Slack main pricing — current Free plan shows up to 10 apps and 90 days of message history. slack.com/pricing
2 Slack Free Plan landing page — current page surfaced with “up to 3 apps.” slack.com/pricing/free
3 Slack feature limitations — current help article specifies 10 third-party/custom app installations. Slack Help
4 HubSpot Free Tools / CRM pricing — current FAQ states up to 2 users and 1,000 contacts. hubspot.com/pricing/crm
5 HubSpot small-business CRM page — current page surfaced with unlimited users and 1 million contacts. HubSpot CRM
6 HubSpot “What is CRM?” page — current page likewise surfaced with unlimited users and up to 1 million contacts. HubSpot
7 Asana Personal plan details — current signups: up to 2 seats; qualifying legacy users: up to 10. Asana Help
8 Asana pricing — current Personal plan and paid-tier pricing. asana.com/pricing
9 Historical Asana Organizations article — old official content still accessible describing free teams up to 15 members. Asana
10 Zapier pricing — current Free plan: 100 tasks/month and unlimited Zap workflows. zapier.com/pricing
11 Zapier plan-change documentation — explains that previous Free/Starter workflow-count limits were removed. Zapier Help
12 ZapierPricing.com — March 2026 page still lists 5 Zaps on Free. zapierpricing.com
13 Toki Zapier beginners guide — published September 1, 2026 and still describes 5 Zaps on Free. withtoki.co.uk
14 Mailchimp pricing/help — current Free plan: 250 contacts, 500 monthly sends, 250 daily. Mailchimp Help
15 Mailchimp pricing — current plan comparison. mailchimp.com/pricing/marketing
16 MailchimpPricing.com — March 2026 independent page captured the older 500-contact / 1,000-send Free plan. mailchimppricing.com
17 EmailCloud — March 2026 article likewise describes 500 contacts / 1,000 sends as current. emailcloud.com
18 Airtable plan overview — current Free plan: 1,000 records/base and 1,000 API calls/workspace/month. Airtable Support
19 Airtable Community historical limit — older indexed answers state 1,200 records/base. Airtable Community
20 Miro pricing — current Free plan details and wording around three editable boards. miro.com/pricing
21 Miro Free Plan help — current board-limit documentation. Miro Help
22 Miro Community, September 2026 — recent report of the live UI blocking creation after three boards. Miro Community
23 Semrush free trial — current standard trial page: 7 days. semrush.com
24 Semrush subscription documentation — distinguishes 7-day standard toolkit trials and selected 14-day add-on access. Semrush Knowledge Base
25 Trello pricing — current Free plan: 10 boards and 10 collaborators per Workspace. trello.com/pricing
26 Trello support — current Free plan and collaborator-limit details. Atlassian Support
27 monday.com Free Plan — current limits: 2 seats, 3 boards, 200 base items with referral expansion. monday.com Support
Research note This report records the public information environment observed during the research window. It is not a permanent pricing guide and does not allege intentional misinformation by any company or publisher. In fast-changing software markets, many contradictions arise from ordinary update lag, legacy entitlements, experiments and distributed content ownership.
From the blog
Latest posts.
No lock-in, just AEO for FREE.
