How to use an AI research assistant for SaaS competitive analysis of pricing and reviews

The best AI research assistant for competitive analysis of SaaS pricing and reviews is the one whose documented controls fit the step you're on, and we found no single tool that covers every step. For pricing pages and feature lists, use an assistant you can limit to the competitor's own website, and for what your team already knows, use one that reads your company's files. Read review complaints yourself, then let the assistant group your notes into themes. Whichever you pick, save a dated copy of every source page first, so your comparison table can cite it and a price change is visible later.

This guide walks through each step, sums up what ChatGPT, Gemini, Perplexity and Glean say about their own research features as of October 2026, and gives you a template you can repeat every quarter. We make Barside, a free desktop app for Windows, Linux and Mac that keeps dated copies of web pages inside your notes. We read the official pages cited for each claim on October 11, 2026 and ran no hands-on tests of their products. Every prompt below is an example to adapt.

Which AI research assistant is best for SaaS competitive analysis of pricing and reviews?

The best AI research assistant for competitive analysis of SaaS pricing and reviews changes from step to step, so match the tool to the job in front of you. The table pairs each step with the control that matters most and a feature its maker documents. Each product does more than one row suggests, and the next section has the detail.

Pricing comes first for a reason. A price belongs to the vendor's own page on a given day, and a figure repeated on a blog or a comparison site can be months old. An assistant that searches the open web may pick up either one, so the site limit matters more for pricing than for any other step.

What Perplexity, ChatGPT, Gemini and Glean say their research features do

If you're choosing an AI research assistant for competitive analysis of SaaS pricing and reviews, the official features of Perplexity, ChatGPT, Gemini and Glean differ mainly in where each one looks and how much say you get over its sources. The table sums up what each company's own page said as of October 2026. We tested none of these features, so read it as a summary of the documentation, and open the linked pages before you rely on a detail.

OpenAI Academy's deep research guide, written for people who use ChatGPT at work, describes a Sites / allowed domains setting that restricts web research to specific domains, which is the control a pricing check needs. The guide says the same setting can instead prioritize those domains while ChatGPT still searches the open web, so choose to restrict when you only want the vendor's own pages. Google's page lists competitive analysis among the uses of Deep Research: a competitor report that cross-references public web data with your internal strategy memos, feature-comparison spreadsheets and team chats.

Perplexity describes its Research mode on its own help page. We give no figures for it here, so read that page for its current limits and plans before you use it for a first scan of who the competitors are. Glean's documentation is written for companies: Deep Research there writes a cited report from your company's sources and the web, takes 5 to 30 minutes, and only reads. One of its sample questions is a competitor comparison built from their blogs, marketing pages and news, and an admin has to turn the feature on before you can run it.

On the Gemini page we read, we found no list of allowed sites, and for Perplexity's app we point you to its help page for the current source controls. With those two, name the competitor's domains in your prompt and check where each citation points, as described in how to restrict AI search to specific websites.

What an AI tool can and can't do in a SaaS competitive analysis

An AI tool for SaaS competitive analysis of pricing and reviews is good at three jobs: finding the right pages, pulling the same fields from each one, and grouping long text into themes. Those are the slow parts of the work, and handing them over is reasonable. Five limits decide how far you can trust the result.

• A citation points at the live page. A live pricing page shows today's prices only. When a competitor changes a plan, the link in your old report shows the new numbers, and the proof of the old ones is gone unless you kept a copy.
• Pricing pages hide detail behind switches. Monthly and yearly billing, currencies and seat counts can sit behind a toggle. Say which billing period you want, and ask for "Not listed" when a value is missing, so the model has no reason to estimate.
• Some prices are not public. A plan marked "Contact sales" has no number to find. Record it as not public, and keep any figure from a forum post in a separate column with its source.
• Models misread tables. A feature tick in the wrong column or a price from the wrong plan reads as confident as a correct one. Check every value against the page before it enters your table.
• Review sites set their own rules. A site's terms can restrict automated collection, which changes how you gather complaints. The next section covers that.

How to gather review complaints within a review site's rules

Review sites decide how their content may be collected, so read a site's terms before you point an agent at it. G2, a site where people review software, is one example. Its Terms of Use, last updated on July 9, 2026, say you may not collect, copy, scrape, store or archive content from the site, including user reviews, through automated means without G2's express prior written consent, whether or not the content is publicly accessible. The same section also bars, without that consent, using content from the site to train, test or improve a machine-learning model or a generative AI system.

A slower method stays on the careful side and gives the assistant better input. It takes five steps.

1. Read recent reviews yourself, in your browser, on the review site. Start with low and middling ratings, where people explain what went wrong.
2. Write one line for each complaint in your own words: the product, the complaint, the plan or company size if the review shows it, the month, and the address of the page.
3. Tag each line as pricing, features, support or reliability, so it can join the comparison later.
4. Give the assistant your lines and ask it to group them. Example prompt: "Below are my notes on 40 recent reviews of one product. Group the complaints into themes, count the notes under each theme, and list the note numbers. Add nothing that is not in my notes."
5. Report counts. "9 of 40 notes mention a seat minimum" is a finding you can defend, while "many users complain" gives a reader nothing to check.

Complaints about seat minimums, paid add-ons and renewal increases deserve a tag of their own, because they can describe costs that a pricing page doesn't show. For the same reason, the template below keeps saved copies for vendors' own pricing and feature pages, and keeps only your own notes for reviews.

A repeatable template: save each pricing page, then fill the table

An AI tool helps most with competitor analysis of SaaS pricing pages and reviews in 2026 when it works from a fixed template, because the same pages, fields and prompts make this quarter's run comparable with the last one. The template has six steps, and it works with any assistant from the tables above.

1. List the pages. For each competitor, write down the exact address of the pricing page, the features page and the changelog or release notes.
2. Save a dated copy of each page before any AI reads it. The copy is your evidence, and the date is part of it.
3. Ask for the same fields from every page. Example prompt: "Read only these pages: [pricing page address], [features page address]. For each plan, write one line with the plan name, the price per user a month on monthly billing, the price per user a month on yearly billing, any seat minimum, the free plan or trial, and the three limits that differ most from the plan below. After each value, give the address you took it from. If a value is not on the page, write Not listed. Do not estimate."
4. Check each value against your saved copy, then enter it in the table with the source address and the date of the copy. A value that fails the check stays out.
5. Add the review themes from your own notes as a second table, with the counts.
6. Repeat on a schedule. Save new copies beside the old ones, and change a row only when the new copy differs.

The finished table looks like the one below. Its rows are made-up examples that show the shape, and none is a real product or a real price.

The last two columns carry the proof. Anyone who doubts a number can open the copy from that day and read it there.

How to run the template in Barside

Barside is a free desktop app for Windows, Linux and Mac, and it brings no AI model of its own: the AI assistant and the agents use the AI account you connect. What it adds to this template is one place where the dated copies, the research notes and the table sit together on your computer.

1. Make a workspace. Download Barside and open it. In the sidebar, under Workspaces, press + (New workspace), type a Workspace name and choose Create workspace.
2. Connect an AI. Open Settings → AI and connect Claude, ChatGPT or OpenRouter, which can search the web in Barside. Gemini can be connected too, but it has no web search in Barside. Under AI spending limit, turn on Limit what Barside's AI spends.
3. Save the copies. Make one note for each competitor. Choose Save a webpage into this note (the download icon in the note toolbar), paste the pricing page's address into Website address or choose Use current browser tab, then choose Save into note. The copy appears under the page's title with Offline copy, the date you saved it and the website's name. Save the features page into the same note.
4. Let the research agent read the pages. Open Agents & approvals → Research and type a Question. Example question: "What does each plan on these pages cost per user a month on monthly and on yearly billing, and what are its main limits?" Add the addresses under Links to read first (up to 5), pick your workspace under Save the note in and press Start research. The research agent writes a note that cites each fact with a number and ends with a numbered list of sources. It can also search DuckDuckGo, so look at its card afterward, which lists every site Barside contacted.
5. Ask wider questions in the assistant. Press Ctrl+Shift+A, type the request and press Run. The assistant can search the web through your AI company and read a web page by its address. It takes up to 40 steps for a request and writes what it finds into a note, with the sources.
6. Build the table. In a new note, type / and choose Database table. Click a column's name to rename it and pick its Type: Text for the plan, Number with the US dollar format for prices, Link for the source page and Date for the day you saved the copy. The guide to database tables in your notes covers sorting, filters and totals.
7. Check, then type. Open the saved copy next to the research note, confirm each value on the copy, and enter it in the table.
8. Set the next check. Open the table's note, press the bell next to its title (Set reminder), pick a date under Remind me and choose Set reminder.

Two details shape how you split the work. The research agent reads up to 5 public pages for a question, with at most two from one site and no PDFs, so two competitors with two pages each fit in one run. A saved copy keeps the look of the page, and the search bar and Library search do not read the text inside it. The research note carries the text you can search, and the copy carries the proof.

Once the table exists, your AI can read it. The assistant and Ask library read a database table as rows of words, so you can ask about it. Example question: "Which plans in the comparison table cost more per user on yearly billing than our Team plan?" The pages the research agent read are saved as bookmarks on its note, with readable copies, so AI search that cites its sources can answer from them later.

Scripts are removed from a saved copy, so a switch between monthly and yearly billing won't flip there. If a pricing page shows one billing period at a time, write the other one into the note by hand from the live page, with the date. Barside saves what a signed-out visitor sees, one page for each save, and tells you why when a page will not save, for example a robot check or a sign-in wall. Saving works the same on Windows, Linux and Mac; in Barside for Mac, Firefox and Zen do not share their address, so paste it there.

How do you see a price change three months later?

To see a price change later, save the same pages again into the same notes. Each copy carries its own date, so the new one sits in the note beside the copy from last quarter, and Open original website takes you to the live page. Update a row only when the two copies differ, and keep the old price in a column of its own, so the table records the change as well as the current state.

To turn the reading step into a routine, choose Save as workflow under the assistant's answer when it finishes a pricing check, type a Workflow name and press Save workflow. Then open Agents & approvals → Workflows, choose Schedule on its card, pick Monthly, tick Ask me to run it on this schedule and press Save schedule. At the planned time the run waits in the Inbox until you press Approve, which is a good moment to save the fresh copies first. The saved workflows guide has the details.

If people outside Barside need to open the evidence, the Internet Archive's Wayback Machine is a public option. Its help page says Save Page Now saves a single page, including its images and CSS, and gives you a permanent address for it, with exceptions for some sites that prohibit crawling. A public copy and a private one serve different readers, and you can keep both.

An example: a quarterly pricing check

As an illustration, picture Priya, who handles product marketing at a small scheduling SaaS and tracks four competitors. On the first Monday of the quarter she saves each competitor's pricing and features pages into four notes, eight copies in all. She runs the research agent twice, with two competitors in each run, and gets two notes with numbered sources.

She checks each price against its copy and types the rows into a database table, with "Not listed" for one plan that only says "Contact sales." From forty lines of her own review notes, the assistant groups the complaints into themes with counts. Three months later she saves the pages again, and one competitor's new copy shows a higher yearly price for its Team plan. When her manager asks when that changed, she opens both dated copies.

Costs, limits and privacy

Barside is free, and it has no AI subscription of its own: the AI company you connect bills your account for what the assistant and the research agent use, including the assistant's web searches. To cap that, open Settings → AI, turn on Limit what Barside's AI spends under AI spending limit, and set an amount from US$0.50 to US$10,000 for a day, a week or a month. Barside warns at 80% and stops its AI at the limit. For flat plans compared with pay as you go, see AI research assistant pricing.

A saved copy is a file in Barside's media folder on your computer, attached to its note, and Barside does not upload it unless you turn on cloud backup or sync with Barside Plus. To make the copy, your computer visits the website once, without your browser's cookies or sign-ins. When the assistant or the research agent works, your request and the pages it reads go to the AI company you connected, whose data policy applies, and the research agent also sends its search to DuckDuckGo. Read more under the AI you connect and your data and how agents use your data.

A competitive analysis often holds your own pricing plans and sales notes. If some of them should stay away from AI, untick those notes or their workspace under Library & briefing → Sources. Excluded items stay out of Library answers and are never given to an agent.

How it compares

A deep research service can read far more in one run than Barside does. Google's page speaks of up to hundreds of websites for Gemini Deep Research, while Barside's research agent reads up to 5 pages. For a wide scan of a market, such a service is the stronger choice. Barside's part is the evidence and the table: dated copies and checked values in one workspace on your computer, using the AI account you already have.

The two work well together. Run the wide scan in a deep research tool, save its report, and keep the saved pages and the checked values in Barside. If you'd rather stay in Claude Desktop or ChatGPT, AI connections let that app search and read the workspaces you choose, read-only by default. For more on choosing tools, see the best AI research tools with citations and AI research assistant team plans.

Works well with