Knowledge Base Software for Small Teams: Costs and Tests (2026)

For a three-person software team, bundled help desk plus help center plans ran from $0 to $165 a month for one product and $89 to $345 for three, on annual list prices checked on 9 November 2026. Docs-only tools add $99 to $249 before you buy a help desk. The spread comes from free tiers and from how many products each plan covers. Test search and AI citations before comparing prices.

Key takeaways

• Price the team you are, not the tool. Bundled plans ran from $0 to $165 for one product and $89 to $345 for three, so product count moves the bill.
• Products are the hidden meter. Zendesk gates multiple help centers by plan, Help Scout and GitBook charge per site, and Convot charges per app.
• If an AI agent reads your help center, price that too. At 500 AI-answered conversations a month the AI line can cost more than the help center itself.
• Test search and citations before price. Drop any tool that cannot find an article by an error code, and if an AI agent will answer customers, any tool whose answers do not show their source.

What does knowledge base software cost a small software team?

The tools in this table cost between $0 and $345 a month for three people, and for several of them the price rises with product count. The table prices three people who edit articles and answer customers, first with one product and then with three. Every price comes from the vendor’s own page on 9 November 2026, in US dollars, before tax, on annual billing unless a column says otherwise.

Monthly cost of a customer help center for three people (annual billing, checked 9 Nov 2026)

Convot data and vendor pages, 9 Nov 2026. The assumption is three people who all edit articles. Where a pricing page does not say how a tool handles several products, the cell says so instead of guessing. Sources are each vendor’s own pricing page.

Three readings of the table matter more than the ranking.

Docs-only tools still need a help desk on top. GitBook, HelpDocs and Helpjuice sell documentation only. A three-person team still needs somewhere to answer the customers who do not find the article, so the real comparison is a bundled help desk against a docs tool plus a help desk.

Two tools list a $0 plan. Help Scout’s Free plan covers up to 5 users and one Docs site, and Convot’s Community plan is free under $1,000 MRR on one app. If you support one product and few contacts, $0 is the honest answer.

Products change the winner. For one product, Convot and Freshdesk sit close together. At three products the bundled options run $89 (Convot Starter with two added apps, 3 seats), $95 (Help Scout Standard with one added Docs site, up to 25 users), $285 (Crisp Essentials) and $345 (Zendesk Suite Professional, needed because Suite Team includes one help center; it also bundles features beyond extra help centers, so read it as an upper comparison). GitBook is $231 before a help desk. The Help Scout figure covers three Docs sites only, with the 2 included inboxes. Convot’s figure gives each app its own help center, and every app lands in the same inbox.

How do multi-product help centers change the bill?

A multi-product help center changes the bill because vendors meter products differently: by plan tier, by site, by workspace or by app. Ask which of those your vendor uses before you count the first invoice.

• By plan tier. Zendesk lists 1 help center on Suite Team and up to 5 on Suite Professional, which is $55 versus $115 per agent a month on annual billing.
• By site. Help Scout includes 1 Docs site on Free, 2 on Standard, 3 on Plus and 5 on Pro, with extra sites at $20 a month on annual billing. GitBook charges $65 per site.
• By workspace. Crisp bills per workspace, so each website you add is another subscription, with a 20 percent discount listed for three workspaces on Essentials and Plus.
• By app. Convot includes 1 app on Starter, 2 on Growth ($79) and 5 on Pro ($149), each with its own help center, on its own domain or under one shared hub.

A team that adds a third product in year two should price that year-two bill now, not the day-one screenshot. The help center for every app page shows how Convot structures several apps under one account.

What does the AI add to the bill?

AI answering adds a second meter that can exceed the help center price, so price it at a volume you expect. The table assumes 500 conversations a month answered by the vendor’s AI agent from your articles, and the units differ, so read the last column before comparing rows.

AI answering cost at 500 conversations a month (list prices, checked 9 Nov 2026)

Zendesk also lists a committed rate for extra resolutions. It showed GBP 1.30 on 9 November, so we use the US$2.00 pay-as-you-go price that the page shows in dollars. The Convot figure treats all 500 as resolved, so it is an upper bound. Six vendors are listed, so these are examples, not the whole market.

None of these numbers says which agent answers better. They say what a wrong guess about volume costs. The comparison of AI support agents that cite their sources covers answer quality and handoff behaviour.

How do you test whether search finds the right article?

Test search with your own tickets, not the vendor demo. Findability is a general problem: in its State of Teams 2024 survey of 5,000 internal knowledge workers, Atlassian reported that 55% found it difficult to track down information despite knowing many people at their organization. Search quality has three measures. Recall asks whether the right article appears in the results. Precision asks how much of what comes back is useful. Mean reciprocal rank, or MRR, asks how high the first useful result sits.

A small ACM study on a simulated food-manufacturer knowledge base reported that hybrid retrieval beat keyword-only and vector-only search, which matches how the two methods behave. Keyword search keeps exact matches for product names, error codes and plan labels. Vector search matches meaning, so a customer who types “my store stopped syncing” still finds “Troubleshoot failed catalog imports”.

Run this before any trial:

1. Write 30 real questions from recent tickets. Include exact identifiers, vague descriptions and questions that belong to different products.
2. Name the expected article first. Then record whether it appears in the top five, so a weak result cannot be rationalized afterward.
3. Misspell the product name once. A tool that ignores typos can frustrate mobile users.
4. Log zero-result searches every week. Separate missing content from content that exists but is not retrieved.

Fix recall first. An article that never reaches the results cannot support an answer.

Does the AI cite the article it used?

Every AI answer needs to link the article it came from, because an answer is only as safe as that article. Also test whether the agent filters by product and hands off when unsure. These controls belong on the scorecard before the first customer sees an AI answer.

1. Source shown and openable. Every answer links the article it came from. Check that the cited passage supports the instruction and not only the topic.
2. Scoped retrieval. The agent filters by product, locale, publication status and date, and keeps internal-only articles out of customer answers. An enterprise retrieval study on arXiv found that metadata-enriched approaches beat content-only baselines, with one setup reaching 82.5 percent precision. That supports testing structure and metadata, not only the model.
3. A handoff threshold. Ask three questions the AI should not answer: a refund for a plan you do not offer, an unsupported feature and a conflicting policy. It should hand off, not invent.
4. An owner and an expiry date per article. Review pricing and setup articles monthly as a starting cadence.
5. A weekly sample. Check 20 AI answers for correctness, citation validity and tone, including ones customers accepted.

Here is what that looks like in our own data. Appointo, one of Sidepanda’s own Shopify apps and not a third-party customer, has an 83-article help center in Convot. In the 30 days to 14 October 2026, Cove AI evaluated 808 customer messages across 230 conversations. It answered 16% directly, drafted a reply for a person to approve on 55%, and escalated 29% because no article passed its confidence threshold.

Convot data, 30 days to 14 Oct 2026. Rates are shares of 808 customer messages, counted per message and not per conversation.

The 29% is the useful number for a buyer. It is an upper bound on content gaps, from one app and one 30-day window, because an escalation can also come from handoff rules. When Cove does answer, it shows the customer which article it used, so a wrong answer can be traced to the article that needs fixing. How Cove AI grounds its answers covers the four source types and the Suggest mode.

Which six criteria should you score before comparing prices?

Score every candidate on the same six criteria, using pass, partial or fail. Pass means the capability works on your content. Partial means you need a workaround. Fail means your team inherits the manual process you meant to replace.

Six criteria to score each knowledge base tool

A vendor demonstration is not a pass. Upload an article with an unusual product name and a real error message, search for both, then misspell the product name and try again.

Localization needs more than a language selector. Verify that visitors receive the right language, can change it themselves and do not see unreviewed machine translations of billing or setup instructions. Integrations deserve the same scrutiny: a REST API and webhooks keep two tools in sync, while a CSV export only supports a one-time move. Map the handoffs with a help desk feature checklist before assuming two connected tools share everything.

Buying rule: Drop any tool that fails search. If an AI agent will answer customers, drop any tool that fails AI grounding too, even if every other row passes.

Keep an exit-route check beside the scorecard. Export an article with images, links and metadata during the trial. A product can pass the publishing test and still leave an expensive cleanup when you move.

How do you move an existing knowledge base?

Move a knowledge base in a controlled order: pick the articles that matter, assign owners, export into a test workspace, redirect old URLs and launch one product first. Moving everything first and cleaning later makes it hard to tell inherited problems from migration damage.

1. Start from the 50 most common questions of the last 90 days. They define the first articles and keep the words customers use.
2. Assign one owner and one expiry date per article. “Support” is not an owner.
3. Export with full history where supported, then import into a test workspace. Compare 10 random articles with the originals for broken images, missing headings and stale links.
4. Redirect old help center URLs one to one. Sending every old URL to the homepage makes customers search again.
5. Launch search and the chat widget for one product first. Review after 30 days and fix the top 10 issues using zero-result searches.

The video below is a further planning resource for the move; we have not reviewed it line by line.

Use a help desk migration checklist to track the operational work alongside content cleanup, and keep the old export until redirects are confirmed. A healthy launch should move toward fewer repeat tickets and a falling zero-result rate in the first month. Treat those as outcomes to measure, not results a migration tool can guarantee. For the article structure itself, see building the knowledge base.

Should a Shopify app team want billing data in the help desk?

Shopify app teams can answer billing questions with plan and charge context beside the conversation, but it needs server-side access and partner permissions, not a browser widget. Shopify documents the Active Subscription and Historical Events APIs, the Partner API subscription object (transaction queries need View financials permission) and usage-based billing. Decide who may see revenue before granting access.

Where does Convot fit?

Convot fits small software teams that run one to several products and want the help center, inbox and AI agent on one bill. Convot makes a competing product, so read this recommendation with that in mind. A larger suite with deeper sales automation or enterprise requirements may suit a funded team better, and staying put can be right when replacing workflows costs more than consolidating tools.

Convot is built by Sidepanda, a small studio, and Shopify app teams are the group we built for. We price by app count with seats on one flat bill. Included seat and app caps still matter, so whole-team pricing is not unlimited capacity.

Convot’s current plan prices

Extra seats cost $10 each and extra apps $20 each on paid plans. Cove AI costs $0.20 per resolved conversation on every paid plan, answers from each app’s own help center and docs, cites its sources and hands off to a human when it is not sure. Agents also see the merchant’s plan, MRR and past billing pulled from the Shopify Partner API.

Crisp imports in one click. Zendesk, Intercom and Freshdesk moves are free and done with the founder, not automatic importers, so ask what moves with you: articles, conversations or both. A lower AI rate does not replace the search, citation and handoff tests above. To see the help center side on its own, open Convot’s help center page. For how large products word and order their front pages, see six FAQ and help pages checked live.

How do you choose with a two-week trial?

Pick the tool that passes search and AI grounding at a cost you can keep for 12 months. A two-week trial should produce evidence for that decision, not a workspace that merely looks finished.

1. Shortlist no more than three tools. Remove candidates that fail your scorecard before configuring them.
2. Load your top 30 articles into each workspace. Use identical content, owners and expiry dates.
3. Run 30 real customer questions. Record whether the right article is in the top five and note wrong-product matches separately.
4. Ask three questions the AI should not answer. Check that it hands off.
5. Export a full copy of your content. Inspect images, formatting and metadata.
6. Calculate the 12-month cost. Use your product count, expected hires and AI volume, with each vendor’s resolution rules.
7. Write one paragraph naming the winner. State why it passed, what compromise you accept and who owns maintenance.

Start the trial this week. Drop any tool that fails the search test, and the handoff test if you will use AI, however low its price. Keep the export test as a condition of purchase.

FAQ

How much does knowledge base software cost?

Knowledge base software costs $0 to $345 a month for a three-person software team on the list prices we checked on 9 November 2026. Free plans start at $0, and paid bundled plans start near $49 to $57 for one product. Docs-only tools run from $99 to $249 and still need a separate inbox.

What is the difference between a knowledge base and a help center?

A knowledge base is the collection of answers your team maintains, and a help center is the public site where customers search and read them. Many teams use the words interchangeably, and in most help desks the knowledge base is what the help center displays.

Is there a free knowledge base tool for a startup?

Two tools in our table are free at small scale. Help Scout’s Free plan covers up to 5 users and one Docs site, and Convot’s Community plan is free under $1,000 MRR on one app. Both have limits, so check which one you would hit first.

Do I need a separate help center for each product?

Products with different customers or setup steps usually benefit from their own help center. One mixed collection makes similar product names look plausible and sends customers to the wrong instructions. Check whether the vendor meters help centers by plan, site, workspace or app, because that decides the third-product bill.

Can AI answer customers from a knowledge base?

AI can answer customers from a knowledge base if it uses only your articles, shows which one it used and hands off when the answer is uncertain. On Appointo’s 83-article help center, Cove escalated 29% of 808 customer messages in 30 days, partly from content gaps and partly from handoff rules.

About the author

Tarang Agarwal is the founder of Convot. Convot is built by Sidepanda, a small studio that runs several Shopify apps of its own, including Appointo, Depo and Panda Bundle, each with its own help center in Convot.