A pricing page answers one question and hides three. It tells you what the first month costs, and it hides whether the meter moves per query, per document, per credit — and whether the people who need it are two or twenty. For image generators that gap is annoying. For research tools it decides the purchase, because the work is bursty: nothing for three weeks, then forty sources in one afternoon.
We keep a directory of 178 AI tools and check every entry against the vendor's own pricing page, with a date. This week we went through the whole Data & Research shelf — 14 tools — plus the two adjacent assistants teams actually use as search ("Perplexity is a research tool, is it not?"), and the two workflow chains those tools sit inside. Every figure below comes from those entries, as of 19 November 2026.
What "price" means in this count
Entry price is the smallest published paid tier, normalized to one month and one seat, excluding taxes and annual discounts. It is a floor, not a forecast: credits, query limits, document caps and overage rates are not in it. A free plan is listed separately, because it changes when you start paying, not what you pay.
The 14 entries are all listed here, each with price, free tier, API availability and the date it was checked.
The shelf in one number
Fourteen tools, twelve of them with a free plan, a median of $15 a month, a floor of $0 and a ceiling of $99 (LLM Socket). The cheap end is not a trial: Semantic Scholar and Gigasheet publish $0, and Semantic Scholar also publishes an API.
Look at the cluster between $11.99 and $12 — Consensus at $11.99, Scite.ai at $12, Elicit at $12. Three tools within one cent of each other, ranked 5th, 6th and 7th cheapest in the category, and they are not the same purchase at all. Scite ships an API. Consensus and Elicit do not. If your reading pipeline needs to be called from somewhere else, two of those three are a dead end at the same price as the one that works — a difference you will not find in any price table, including ours, unless you read the API line.
The filing problem: where Perplexity actually sits
Perplexity costs $20 a month and it is the tool most teams name first when they say "AI research". In our catalogue it is not in Data & Research. It sits in Assistants & LLMs — 17 tools, median entry price $10, floor $0 (You.com), ceiling $50 (Cohere Coral). At $20 it is the 14th cheapest of 17 — 13 tools cost less, three cost more — while on the research shelf $20 sits mid-table, with nine of fourteen tools cheaper and a tie at Julius AI and SciSpace.
That is not a trick of taxonomy, it is the whole argument. The same $20 reads as "mid-priced" or "expensive" depending on which shelf you compare it on, and the shelf is chosen by the vendor's positioning, not by your job. The direct comparison that matters is Perplexity vs You.com: $20 a seat against $0 a seat, both with an API. The full alternative list is here.
Seats: where a $20 decision becomes a $400 one
Every tool on this shelf bills per seat, linearly. Nobody offers a research pool, a shared credit balance, or a five-for-three. So the price question is really a headcount question.
At one seat the spread is $0 to $20 — the price of a lunch. At twenty seats it is $0 to $400 a month, and the four numbers behave nothing alike: Semantic Scholar and You.com publish $0 at 1, 5 and 20 seats; Consensus goes from $11.99 to $59.95 to $239.80; Perplexity from $20 to $100 to $400.
Two things follow, and they pull in opposite directions.
The first: for the two $0 tools, "how many seats" is a non-question, which is exactly why they get used as a default and why the paid tool has to justify itself on output rather than on access. The second: research access is rarely uniform across a team. One analyst reads forty papers; five colleagues read one summary each. Billing all six the same way is the expensive mistake in this category — and it is the one the pricing model encourages, because there is no way to buy a single heavy seat and five light ones.
The 1/5/20 seat counts are scenario assumptions in our data, not vendor tiers. The per-seat prices are vendor figures.
Two stacks, four steps, two totals
Nobody buys research tools one at a time either. We model them as chains: each step of a job, the first pick, and the cheapest tool that can do the same step.
The market research chain is the clearest example: sources, extraction, analysis, report. First picks are Perplexity ($20), ChatPDF ($19.99), Julius AI ($20) and Gamma ($10) — $69.99 a month. Swap only the steps that have a cheaper equivalent and the same four steps cost $51.98: Consensus at $11.99 instead of Perplexity (-$8.01), Humata AI at $9.99 instead of ChatPDF (-$10). Julius and Gamma have no cheaper swap in our data.
$18.01 a month, 26% of the stack, for the same four steps in the same order. The full step-by-step page is AI Market Research & Competitor Analysis.
The second chain shows the same effect on a smaller base: Julius AI ($20) + Rytr ($9) + Winston AI ($12) for a published B2B whitepaper totals $41, and drops to $33.99 with HubSpot Breeze AI instead of Julius and GPTHuman instead of Winston — $7.01 less, one workflow page away. And in a third, Perplexity is one of three tools in a $71-a-month SEO pipeline, where it is 28% of the total; swapping it for You.com at $0 takes that stack to $51.
Three chains, three different answers to "what does an AI research budget cost" — $33.99, $51.98, $71 — and none of them is the sum of the tool prices on someone's wishlist, because the same tool appears in several chains at once.
What the numbers cannot tell you
Three things are missing by construction, and we mark them instead of smoothing them over.
There is no comparable unit. None of the 14 entries publishes a price per query, per document or per answer in a unit that reads across tools. One sells credits, one sells documents per month, one meters "deep research" runs separately. That is why the catalogue carries the monthly entry price per seat and not a cost per question. Any "cost per research answer" figure you see is built on an assumption the seller did not publish.
Free limits are not normalized. 12 of 14 tools have a free plan; what each one carries is not in our data, only that it exists. The only way to find the wall is to run a real question end to end for nothing.
The entries move. Each row carries its own check date. Two weeks between two checks is enough for a price change in this category, so a table without dates is a table you should not budget from.
The order that saves the money
Four steps, in this order, and the expensive mistake usually disappears before step four.
- Run one real question on free tiers. Not a demo question — the one that is waiting on your desk. Twelve of fourteen tools let you do this.
- Find the step that fails, not the cheapest tool. Most research stacks fail at extraction (PDFs, tables, citations), not at search. That is where the paid seat belongs.
- Buy seats for the people who hit the wall. Not for the team, not for the org chart. The billing is linear, so every seat you add is the full price of the tool.
- Check the API line before you fall in love. At $12, Scite has one and Elicit does not; at $0, Semantic Scholar does. If the tool will ever be called from a script, that single line outranks every feature on the page.
The category is unusual in one respect: because 12 of 14 entries start at $0, you can run the entire evaluation before paying anything, and the only genuinely irreversible decision is the seat count. Which is, of course, the one the price list does not discuss.
All prices: published entry tiers, per month and per seat, excluding taxes and annual discounts, checked 19 November 2026. Seat counts are scenario assumptions. The full catalogue with check dates is at ai-saasradar.com/tools.



Top comments (0)