Because nobody in your business owns the job of looking, and until a few days ago the looking cost more than most owners could justify. This article answers one question: what does a standing weekly competitive research routine actually cost to run in October 2026, and how do you set one up without hiring anybody?
"I Found Out From a Customer"
That is the sentence that stings.
A buyer mentions, almost kindly, that the shop across town dropped their price, or added the service you have been meaning to add for two years. You nod. You finish the conversation. Then you sit in the truck, or stare at the screen, and feel something that has nothing to do with pricing.
You feel behind. Not behind on a task. Behind on your own market, the one you have worked in for twenty years.
Here is the direct answer to the question in the title. The reason you did not know is almost never that you lack the skill to find out. It is that finding out took a block of time you did not have, and nobody owned the assignment. That specific constraint changed last week, and it changed quietly, because it arrived as a row on a developer pricing page.
DeepSeek opened the search system that powers its own app to outside developers. Its published rate is $0.75 per 1,000 successful requests for Search, and the same for Fetch, which pulls the content of a page (DeepSeek API pricing page, checked October 19, 2026). Failed requests, empty results and timeouts are not billed.
Ten thousand web searches now carries a published price of seven dollars and fifty cents.
That is the whole argument. When a search costs less than a tenth of a cent, checking what your five closest competitors changed, every week, indefinitely, stops being a project you schedule and becomes a background process you forget is running.
Competitive research was never limited by the searching. It was limited by the fact that a human being had to remember to look, and that human being was you, and you were booked solid.
Key Takeaways
Renders as a shaded callout box.
- DeepSeek lists its Search and Fetch APIs at $0.75 per 1,000 successful requests, the cheapest published rate among search APIs agent builders use.
- Per-request billing has not started, so that is a published price, not an invoice anybody has paid. Test it small.
- Research since 2000 shows small-firm owners scan their competitive environment quarterly to annually, because operations leave no room.
- The expensive part was never the search. It was the commitment to look, which software can now hold for you.
- Define the decision before you automate the gathering.
The Problem Is Not That You Do Not Care
Competition is not a blind spot for small business owners. It is a stressor they name out loud.
In U.S. Bank's 2026 Small Business Perspective survey, 82 percent of owners reported ongoing strain from competition, behind only the general economic environment at 90 percent and inflation at 88 percent. That survey ran February 27 to March 17, 2026, with 1,000 U.S. small business owners between two and ninety-nine employees and a margin of error of plus or minus 3.1 percent.
So the awareness is there. The structure is not.
Here is the part that stops people, and it is worth naming directly, because it is not laziness and it is not ignorance. It is a sequencing problem. Competitive research produces no revenue on the day you do it. Invoicing does. Payroll does. The customer on hold does. So the research slides, and it slides again, and eventually it only happens when something forces it: a lost bid, a renewal conversation that goes sideways, a bank asking for a plan.
Then it gets done in a panic, by you, at eleven at night, as a one-time project. You pull up six competitor websites, take some notes, feel briefly informed, and let the file go stale.
There is a second problem underneath the first, and it is quieter. A lot of experienced owners have started to suspect that their market knowledge, the thing they have been paid for, has aged out from under them. That is not overwhelm. That is confidence collapse, and it tends to make people look less, not more, because looking confirms the fear.
I want to be careful not to minimize either problem. The time scarcity is real. The fear is reasonable. Thirty years of selling taught me that people do not act on information they are afraid to hear, which means a competitive research habit that only delivers bad news will die in three weeks.
But the reframe is available now in a way it was not available in January. The constraint everyone assumed was money, or technical skill, or access to data, was actually attention. And attention is the one thing you can now hand to a machine without hiring anybody, because the underlying plumbing got absurdly cheap.
The Evidence
Four things, each with a named source and a date.
One. The price of a web search collapsed, and you can check it yourself.
DeepSeek's pricing page lists Search and Fetch at $0.75 per 1,000 successful requests, drawing from the same balance as token usage. Per-request billing starts October 31, 2026 at 16:00 UTC, midnight Beijing time on November 1, and calls through DeepSeek Harness, Claude Code or Codex are free until then. The same release shipped V4.1-Flash-1017, an updated model aimed at faster agent and coding work, with open weights (DeepSeek release notes, October 17, 2026).
Here is how that compares. Each figure came off the vendor's own pricing page on October 19, 2026.
| Service | Published price | Unit |
|---|---|---|
| DeepSeek Search | $0.75 | per 1,000 successful requests |
| DeepSeek Fetch | $0.75 | per 1,000 successful requests |
| Exa Instant Search | $4 | per 1,000 requests, up to 10 results |
| Brave Search API | $5 | per 1,000 requests, plus $5 free credits monthly |
| Exa Fast / Auto Search | $7 | per 1,000 requests |
| Tavily basic search | $8 | per 1,000 ($0.008 per credit pay-as-you-go; a basic search uses one credit) |
| Exa Deep Search | $12 to $15 | per 1,000 requests |
| SerpApi Starter | $25 | per month for 1,000 searches |
Two caveats. Third-party compilations put Perplexity's Sonar per-request search fee between $5 and $12 per 1,000, which I could not confirm on Perplexity's own page. Serper advertises rates from $0.30 per 1,000, though reported pack pricing suggests that floor applies only at high volume. Prices move. Check before you budget.
Two. Market knowledge correlates with performance, and the number is not small.
Ahmet Kirca, Satish Jayachandran and William Bearden's 2005 meta-analysis in the Journal of Marketing pooled the empirical literature on market orientation. The corrected mean correlation with firm performance was r = .32, which they categorized as above medium by Cohen's standard. It ran lower in service firms, r = .26, than in manufacturing, r = .37.
Plain language: firms that systematically gather and act on information about customers and competitors perform better. Not dramatically better. Reliably better, across dozens of studies.
Three. Small firms have been checking roughly quarterly to annually, and the reason is time.
Reichel Beal's study of small manufacturers, in the Journal of Small Business Management in 2000, measured how often CEOs scanned five parts of their task environment: competitors, customers, technology, suppliers and the firm itself. The mean frequency landed between quarterly and annual. Beal's reading was that owner-managers, tied up in daily operations, did not have time for more.
Azhdar Karami's 2008 survey of 132 CEOs at UK electronics manufacturers found the same gap from the other side. One hundred seven, 81 percent, called a formal scanning system essential to firm performance. Only ten said it was not. High performers leaned on formal, coordinated scanning; low performers on informal, low-frequency looking.
Owners already believe monitoring matters. They cannot personally sustain it.
Four. Contrast that with people who do this for a living.
In the Competitive Intelligence Alliance's 2021 Trends Report, produced with Klue, 69.4 percent of surveyed competitive intelligence and product marketing professionals monitored competitors at least monthly, and the most common frequency for light spot-checks, 23.6 percent, was a couple of times per week.
That is a sample of enterprise specialists, not small business owners, so do not read it as a benchmark you are failing. Read it as the cadence professionals consider normal when somebody is paid to hold the assignment. The cadence was never the hard part. The salary was.
What a Standing Research Routine Actually Looks Like
I use the conductor and the orchestra to describe this, because it is the most honest picture I know of how AI works in a small business.
The orchestra plays faster than any one musician, in more registers, without tiring. It cannot decide what the piece means, or hear that the second movement is dragging, or read the room. You do that. AI amplifies human skill. It does not replace it, and anybody selling you the opposite is selling you something.
Applied here: the machine goes and looks. You decide what the looking means.
A standing routine has four parts. The first is the one people skip.
A named decision. Not "monitor competitors." A decision you will actually make with the output. Do I hold my price this quarter. Do I add the service they just launched. Do I change the three objections my proposals answer. If you cannot name the decision, the routine will produce a document nobody opens, and you will conclude that AI research does not work, when what did not work was the absence of a question.
A short, fixed watch list. Five competitors, not twenty. The five you actually lose deals to. Plus two or three sources that move your market: a regulator, a supplier, a trade publication.
A fixed cadence and a fixed format. Weekly, same day, same shape. The format matters more than the frequency, because a consistent format is what lets you spot change. Something changed this week, here is what, here is what it suggests, here is what I recommend you decide.
A human review step. Fifteen minutes. You read it. You mark one thing to act on, or you mark it "no action," which is a real answer and should be written down as one.
What it should produce is a decision log, not an archive. Twelve weeks in, the value is not the fifty-two reports. It is the trail of nine decisions you made with your eyes open, and the pattern you can now see in your own market that you could not see from the inside of any single week.
And this is where I need to say the thing I say about everything: validate before automate. Run this by hand for three weeks before you schedule it. Do the searches yourself, or in a chat window, and read what comes back. You are checking whether the output is accurate about your market, whether the format gives you something you can act on, and whether you actually read it. If you will not read it by hand, scheduling it will not fix that. It will just make it expensive, in attention rather than dollars.
The plumbing side got easier the same week the price dropped. CrawlForge MCP launched on Product Hunt today, October 19, 2026, from maker Simon Lacey. It is an open-source MCP server that hands an AI agent 31 web tools, scrape, search, crawl, structured extraction and persistent browser sessions, through one install command, with 1,000 free credits to start. Cheap search is one half. A standard way to hand that search to an agent is the other half. Both landed in the same week, which is why this is worth your attention now rather than next year.
Seven Steps to Set This Up This Week
1. Write down the decision first. One sentence, on paper. "By January I need to decide whether to hold my current pricing." Everything downstream serves that sentence. If you skip this step, skip the rest too, because you will build a machine that produces homework.
2. Build the watch list. Five competitors by name and URL. Two or three market sources. Put it in one file you can hand to an AI tool in a single paste. Name it something you will find again.
3. Run week one entirely by hand. Open a chat with whichever AI tool you already pay for, paste the watch list, and use this prompt. Read the output critically. You are validating the format before you automate it.
[The Job]
Analyze what changed this week for a specific set of competitors and summarize it for a decision.
This is for: the owner of [YOUR BUSINESS TYPE] in [YOUR MARKET AREA], who has 15 minutes to read this.
It matters because: I need to decide [THE SPECIFIC DECISION] by [DATE], and I want the decision made on current facts rather than on what I remember from last year.
[The Background]
Here is what you need to know: My five closest competitors are [COMPETITOR 1, URL], [COMPETITOR 2, URL], [COMPETITOR 3, URL], [COMPETITOR 4, URL], [COMPETITOR 5, URL]. My own positioning is [YOUR POSITIONING IN ONE SENTENCE]. I typically lose deals to them on [THE REASON YOU LOSE]. Check their public pages for changes to pricing, service lines, staffing, locations, promotions and claims. Also check [TRADE SOURCE] and [REGULATOR OR SUPPLIER] for anything that affects my market.
Do not use: paid-review sites, aggregator pages, or any claim you cannot tie to a source I can open. Do not speculate about their internal strategy. Do not pad the summary with background I already know about my own industry.
[The Deliverable]
Return: a one-page brief, under 400 words, in four sections. What changed. What it probably means. What I should consider deciding. What you could not verify.
Must include: a source link for each claim, and an explicit "no change detected" line for any competitor where nothing moved.
Optimize for: accuracy.
[The Questions]
Ask me any questions you have.
4. Fix the format, then fix it again. Week one will produce something close but wrong. Too long, too vague, or missing the one thing you care about. Edit the prompt, not your expectations. Two or three passes gets you to a brief you will read.
5. Add the plumbing. Once the format holds, connect your AI tool to live web access through an MCP server such as CrawlForge, or whichever your tool already supports, and point the model at a search API. DeepSeek's Search and Fetch endpoints at $0.75 per 1,000 are the cheapest published option today. Start with a hundred requests and read your actual bill before you scale.
6. Schedule it, then audit it monthly. Claude Code and Claude on the web both support scheduled tasks that run a saved prompt on a cadence; the official documentation at code.claude.com covers both the session-scoped and cloud versions. Set it weekly. Then once a month, spot-check two or three claims against the source yourself. Automation drifts. Unaudited automation drifts quietly.
7. Close the loop with a decision, every time. Fifteen minutes, same slot each week. Use this prompt on the accumulated briefs when the decision date arrives.
[The Job]
Decide between the options in front of me, using the competitive briefs I have collected.
This is for: me, the owner, making a call I will be accountable for.
It matters because: [THE SPECIFIC DECISION] is due by [DATE], and I would rather make it from the record than from my gut on the day.
[The Background]
Here is what you need to know: I am pasting [NUMBER] weekly competitive briefs covering [DATE RANGE] below. My options are [OPTION A], [OPTION B] and [OPTION C]. My constraints are [BUDGET, CAPACITY, TIMING]. My current position is [WHERE YOU STAND TODAY].
Do not use: anything not present in the briefs below or in my stated constraints. Do not invent market size figures, growth rates or competitor revenue. If the briefs do not support a recommendation, say so.
Briefs: [PASTE BRIEFS]
[The Deliverable]
Return: a recommendation under 300 words, with the two strongest reasons for it, the strongest argument against it, and the one piece of information that would change the answer.
Must include: a plain statement of how confident you are and why.
Optimize for: accuracy.
Written from the perspective of a skeptical advisor speaking to an owner who does not want to be flattered.
[The Questions]
Ask me any questions you have.
Frequently Asked Questions
Is $0.75 per 1,000 searches a real price I can count on?
It is a published price, not an invoice anybody has paid yet. DeepSeek's per-request billing for Search and Fetch starts October 31, 2026 at 16:00 UTC. Until then, calls through DeepSeek Harness, Claude Code and Codex are free. Run a hundred requests and read your first real statement before you plan around the number.
Do I need to be a developer to do this?
No for the manual version, which is most of the value. Pasting a watch list into a chat tool and asking for a weekly brief requires no code. The automated version means installing an MCP server and connecting an API key, roughly the difficulty of setting up an email client. If that is past your comfort line, run it manually.
How is this different from asking ChatGPT or Claude to search for me?
For one-off questions, mostly it is not. The difference is cadence and format. An ad hoc search answers today's question and leaves no record. A standing routine with a fixed format produces a comparable series, which is what lets you see change rather than status. Cheap search makes the recurring version affordable.
Is it legal to monitor my competitors' websites this way?
Reading public web pages is generally lawful in the United States, and the Ninth Circuit's LinkedIn ruling held that scraping publicly accessible profiles did not violate the Computer Fraud and Abuse Act. Risk comes from method and use: gated areas, terms of service, server load, personal data. This is not legal advice. Ask your attorney.
What should a small business actually budget for this?
At five competitors checked weekly, roughly 2,000 billed search and fetch requests a year, the search line runs under two dollars on DeepSeek's published rate by my arithmetic. Model tokens to read and summarize results are separate and will likely exceed the search cost. Compare that to competitive intelligence consulting, where reported project minimums commonly start around $10,000.
The Looking Was Never the Hard Part
Go back to the truck. The customer who told you something you should have known.
What happened there was not a failure of expertise. Your expertise is intact. You knew exactly what that price change meant the second you heard it, which is precisely the point. The analysis was never your weak spot. The surveillance was, because surveillance is boring and relentless and you have a business to run.
That is the job that just got cheap. Not the judgment. The watching.
So here is the direct version, and I will not dress it up. If you are an experienced operator who has started to quietly wonder whether your market knowledge has gone stale, the fix is not a course and it is not a mindset shift. It is a recurring calendar entry and a prompt. Five competitors. One page. Same day each week. A decision at the end of it, even when the decision is "no change."
Do it by hand for three weeks first. Validate before automate. If you will not read it when it is free and manual, you will not read it when it is scheduled and costs a dollar.
The constraint on knowing your own market was never the searching. It was that somebody had to remember to look. That somebody no longer has to be you.
About the author
Jonathan Mast is the founder of White Beard Strategies LLC, an AI education, coaching and training company he started in June 2023. Before that he built and exited Valorous Circle, a digital marketing agency he grew to seven figures, following a thirty-year career in consultative selling. He serves a free Facebook community of more than 500,000 members called AI Prompts for Entrepreneurs, along with the paid White Beard Strategies AI Insiders(TM) membership. His work is practical, anti-hype AI education for entrepreneurs.
Sources
- DeepSeek API Docs, Models and Pricing (Search and Fetch at $0.75 per 1,000 successful requests; billing start date; deduction rules), checked October 19, 2026
- DeepSeek-V4.1-Flash-1017 Release, October 17, 2026 (Search and Fetch opened to all; billing starts October 31, 2026 at 16:00 UTC; open weights)
- DeepSeek Search and Fetch API guide
- Tavily Pricing ($0.008 per credit pay-as-you-go), checked October 19, 2026
- Brave Search API (Search at $5 per 1,000 requests; Answers at $4 per 1,000), checked October 19, 2026
- Exa API Pricing (Instant Search $4, Fast/Auto $7, Deep Search $12 to $15, Monitors $15, all per 1,000 requests), checked October 19, 2026
- SerpApi Plans and Pricing (Starter $25 per month for 1,000 searches), checked October 19, 2026
- Serper (advertises rates from $0.30 per 1,000 queries), checked October 19, 2026
- U.S. Bank 2026 Small Business Perspective survey, released June 22, 2026; n=1,000 U.S. small business owners, fielded February 27 to March 17, 2026, margin of error plus or minus 3.1 percent
- Ahmet H. Kirca, Satish Jayachandran and William O. Bearden, "Market Orientation: A Meta-Analytic Review and Assessment of Its Antecedents and Impact on Performance," Journal of Marketing 69(2), April 2005, pp. 24-41
- Reichel M. Beal, "Competing Effectively: Environmental Scanning, Competitive Strategy, and Organizational Performance in Small Manufacturing Firms," Journal of Small Business Management 38(1), 2000, pp. 27-47, as reported in Azhdar Karami, "An investigation on environmental scanning and growth strategy in high tech small and medium sized enterprises", High Technology Small Firms Conference, University of Twente, May 2008
- Competitive Intelligence Alliance and Klue, Competitive Intelligence Trends Report 2021 (69.4 percent monitor competitors at least monthly; 23.6 percent spot-check a couple of times per week)
- CrawlForge MCP on Product Hunt, launched October 19, 2026, by Simon Lacey; source on GitHub
- Claude Code scheduled tasks documentation and web scheduled tasks
- hiQ Labs v. LinkedIn, Ninth Circuit, on scraping publicly accessible profiles and the CFAA (context on scraping and trade secret risk)
- userintuition.ai, competitive intelligence pricing overview (reported consulting project minimums around $10,000, based on Clutch data)
