Which AI Consulting Company Should You Choose?

It's worth it when you're buying measurement and a system your own team keeps running after the invoice clears. It's usually not worth it when you're buying content volume. That distinction decides the answer, and almost nothing else does.

Here's what makes this question different from the general one about AI consultants. You are almost certainly already doing AI marketing. Look at the Federal Reserve's 2025 Small Business Credit Survey. Among small employer firms using AI at all, the most common application was writing or marketing, named by 83% of them (Federal Reserve Banks, 2026 Report on Employer Firms, March 2026). Someone in your office is already pasting your service descriptions into a chatbot. So a consultant isn't introducing AI to your marketing. They're being hired to formalize something that's already happening, unmanaged, for free.

That can be worth real money. It can also be the easiest thing in the world to sell and the hardest thing to verify. We build AI into operations-heavy businesses for a living, and marketing is not usually where we start. This guide explains why, and what would have to be true for us to tell you it's the right buy.

Key Takeaways

• Among small businesses using AI, 83% apply it to writing or marketing — the most common use by a wide margin (Federal Reserve, 2026).
• Only 7% of small-business AI users have fully integrated AI into their processes. Half are still experimenting (Federal Reserve, 2026).
• Generative AI content features are already bundled into software you may pay for, including Mailchimp Standard at $20/month.
• When Google shows an AI summary, users click a traditional result on 8% of visits versus 15% without one (Pew Research Center, 2025). Content volume is a weaker bet than it was.
• The buy is worth it if it comes with a measurement baseline. Without one, you cannot tell whether it worked, and neither can they.

Table of Contents

• The Short Answer: Two Very Different Purchases
• You're Already Doing AI Marketing
• Two Federal Surveys, Two Very Different Numbers
• Check What Your Software Already Includes
• The Deliverable Most Often Sold Is the One Losing Ground
• Can You Measure It at Your Volume?
• What You're Probably Spending on Marketing Now
• How to Tell If They're Overselling the AI
• When Marketing Isn't Where the Money Leaks
• Eight Questions to Ask Before You Sign
• When to Walk Away
• Your Next Step
• Frequently Asked Questions

The Short Answer: Two Very Different Purchases

AI marketing consulting is worth it when you're buying a measured baseline, a system, and someone accountable to a number. It's rarely worth it when you're buying output — posts, emails, ad variants — because output is now the cheapest thing in the category. Find your row.

Four of those seven rows say no. That ratio isn't cynicism about the category. It's what happens when a capability gets bundled into software faster than the services market around it adjusts. The general version of this question covers AI consultants of every kind, and we work through it in is it worth hiring an AI consultant for a small business. This page is only about marketing, because marketing is genuinely different.

You're Already Doing AI Marketing

Marketing is where small businesses put AI first, and it isn't close. The Federal Reserve Banks surveyed 6,525 small employer firms with 1 to 499 employees between September and November 2025. Among firms using AI, 83% used it for writing or marketing. Individual productivity came second at 61%. Planning or analysis third, at 51% (Federal Reserve Banks, 2026).

Now the number that matters more. Of the firms using AI, about half described themselves as experimenting. Another 44% had partially integrated it. Just 7% — roughly one in fourteen — had fully integrated AI into the processes their business actually runs on (Federal Reserve Banks, 2026).

Read those two findings together and the real market becomes visible. Marketing is where AI shows up first because it has the lowest cost of failure. A bad email draft costs nothing. A bad estimate costs a job. So the writing gets automated early, informally, by whoever is comfortable with a chatbot — and it stays informal. Nobody owns it. Nothing gets measured. That's the gap a good consultant closes, and it's a legitimate thing to sell. Just notice what it is. You're not buying capability. You're buying management of a capability you already have.

Two Federal Surveys, Two Very Different Numbers

If you compare two vendor pitches, you may see two very different adoption statistics for the same year. Both can be from the U.S. government. Both can be right.

The Census Bureau's Business Trends and Outlook Survey put it at 18%. That's the share of firms using AI in at least one business function between November 2025 and January 2026, rising to 32% weighted by employment (Bonney and others, The Microstructure of AI Diffusion, NBER Working Paper 35141 / Census CES-WP-26-25, April 2026). The Federal Reserve put it at 46%. The surveys disagree by a factor of more than two.

They're measuring different things. Census asks whether the firm deploys AI in a business function — an institutional decision, made once, at the top. The Fed asks whether "the business or its employees currently use AI," which counts your estimator quietly using a chatbot on his phone. Neither is wrong. But one describes a system and the other describes a habit.

Why this matters to you as a buyer: a consultant quoting the 46% is describing a world where you're behind and need to catch up. A consultant quoting the 18% is describing a world where you're early and need to move. The honest reading is that adoption is widespread and shallow. The scarce thing is not access to AI. It's a marketing process that survives the person who set it up leaving.

Check What Your Software Already Includes

Before you price a consultant, spend twenty minutes on your own invoices. Generative AI content features are now included in ordinary small-business subscriptions, at no additional cost, on plans many operations already pay for.

Mailchimp lists "Generative AI Features" as a no-additional-cost add-on on its Standard plan at $20 per month, and on Premium. A subject line helper is included on every tier, the free one included. Semrush folds its AI search visibility and AI sentiment tools into base plans with no AI-specific upcharge, and Klaviyo puts an AI subject line generator in its free plan. Jobber, which a lot of field-service operations already run, includes AI chat and AI voice on every plan. None of that is an upsell.

The exception is worth knowing, because it's where the real money is. Agent-tier AI is metered, not bundled. HubSpot runs AI on a credit system. Starter gets 500 credits a month at $7 per seat; Professional gets 3,000 at $800 a month. A single customer-agent conversation consumes 50. Jobber charges $29 a month for an AI receptionist and $99 a month for its marketing suite.

That last row is the whole argument. Software vendors will keep giving away generation, because generation is what makes their seats stickier. None of them will tell you whether your marketing is working, which workflow to change, or who owns it on the Tuesday after go-live. So buy the row nobody is bundling: the measurement, the process design, and the answer to who owns this in month four.

The Deliverable Most Often Sold Is the One Losing Ground

The standard AI marketing engagement produces more content. That bet is weaker in 2026 than it was in 2023, and there's now real measurement behind the claim.

Pew Research Center recruited 900 U.S. adults onto a browsing-tracking panel. It observed 68,879 Google searches in March 2025, of which 12,593 produced an AI summary. When a summary appeared, users clicked a traditional search result on 8% of visits. Without one, 15% — nearly twice as often. They clicked a link inside the summary itself on just 1% of visits. They also abandoned the session entirely more often: 26% of pages with a summary versus 16% without (Pew Research Center, July 2025).

Two details sharpen this from a scare stat into something you can act on. First, the exposure is concentrated exactly where informational content lives. Pew found summaries on 8% of one-to-two-word searches. On searches of ten words or more, 53%. On searches beginning with a question word, 60%. So "how much does a new furnace cost" draws a summary and loses the click. "Plumber near me" largely doesn't. The blog-post deliverable is the most exposed thing on the menu; your service pages and local listings are the least.

Second, supply commoditized at the same time. Graphite sampled 43,000 URLs from CommonCrawl. AI-generated articles passed human-written articles in volume around November 2024. And those AI-written articles "largely do not appear in Google and ChatGPT" (Graphite, October 2025). Demand for the click fell while supply of the article exploded. Demand for the click fell at the same moment the supply of the article exploded, and both of those trends point the same way for anyone selling content by the unit.

One honest caveat, and it belongs here rather than in a footnote. Google disputes this framing. It argues that AI features create new opportunities for content and businesses to be discovered, and that publisher traffic-loss claims rest on incomplete data. Pew's comparison isn't a controlled experiment either. Ahrefs found position-one click-through on summary-triggering keywords fell from 0.073 to 0.016 between December 2023 and December 2025, and describes its own result as a correlation (Ahrefs, February 2026). Treat the direction as well evidenced and the magnitude as contested. Then ask the consultant which number they're assuming.

Can You Measure It at Your Volume?

Probably not the way a proposal implies. Proving a marketing lift statistically takes volume that most small operations simply don't have, and the arithmetic is published by the federal government. The National Institute of Standards and Technology publishes the arithmetic. Its statistics handbook works a sample-size example for exactly this kind of test. Detecting a jump in a conversion rate from 10% to 20% takes at least 112 attempts per version, at 5% significance and 90% power (NIST/SEMATECH, e-Handbook of Statistical Methods).

But nobody sells you a doubling. Proposals promise improvements in the tens of percent, and a smaller effect is a much harder thing to prove. Run the same formula at the same settings and the requirement climbs fast.

The first row is NIST's published example. The other three are our own arithmetic using the handbook's formula at identical parameters, shown so you can check them. The reason for the curve is that the required sample scales with the square of the precision you're after. Halve the effect you want to detect and you quadruple the traffic you need.

Now put real volumes against that. In NAHB's member census, the median builder member started six homes in 2024. That figure hasn't moved since 2021. Median revenue was $3.7 million, on six employees (NAHB, August 2025). The typical REALTOR® closed nine transaction sides in 2025 (NAR, June 2026). Inquiries run higher than closed jobs, of course. They don't run three orders of magnitude higher.

Practitioners who run these tests for a living put the viability floor at a few thousand monthly visits and roughly 100 conversions per version (Seer Interactive, 2022). Most operations reading this clear neither bar.

There's a subtler trap waiting for anyone who tries anyway. Checking a running test repeatedly and stopping when it looks good inflates false positives badly. With a 50% baseline, continuous monitoring against a 5% significance threshold produces a 26.1% false-positive rate (Evan Miller, 2010). So the small-volume test that finally shows a win is the one most likely to be noise. A consultant who reports weekly lift numbers on 40 leads a month is reporting weather rather than climate, and you're being asked to renew on it.

Here's the useful conclusion, and it's not "give up on measurement." Stop trying to measure rate differences. Start counting events. How many of last month's inquiries got a response the same day? What's the median hours from inquiry to quote sent? How many quotes never got a follow-up at all? Those are countable at forty leads a month, they're unambiguous, and every one of them moves revenue you can actually see. Buy work that changes numbers like those. The deeper version of this argument is in how to choose an AI consulting service for business ROI.

What You're Probably Spending on Marketing Now

You need a denominator before a fee means anything. The CMO Survey publishes marketing spend as a share of revenue, broken out by company size. It's run by Professor Christine Moorman at Duke's Fuqua School of Business, and its 35th edition surveyed 308 U.S. marketing leaders in January 2026, 97% of them VP-level or above. Here's the part worth reading carefully.

Look at the standard deviation column before you use any of these numbers. It's roughly as large as the mean in every small-firm row. That means the spread between comparable companies is enormous and the average describes almost nobody. Two caveats compound it. The cells hold only 19 to 30 firms each. And every respondent is a senior marketing leader, so a sub-50-employee company in this sample employs a VP of marketing. A 30-person contractor does not.

So don't plan against the benchmark. Use it for one thing: a sanity check on scale. Say you run a $6 million operation. Marketing might plausibly be a few hundred thousand dollars a year. A $30,000 engagement is a real fraction of that, and it should arrive with a number attached. If a proposal costs more than a quarter of your annual marketing spend and comes with no baseline, the pricing is inverted. For what the market actually charges, see how much AI consultants charge.

How to Tell If They're Overselling the AI

This is not a hypothetical risk, and the enforcement record is specific enough to use as a checklist. In September 2024 the Federal Trade Commission announced Operation AI Comply, five actions against companies using AI claims deceptively. Then-Chair Lina Khan's framing was blunt: "there is no AI exemption from the laws on the books" (FTC, September 2024).

Three of the resulting cases map directly onto marketing services.

Accuracy claims need evidence. Workado marketed its AI content detector as "98 percent" accurate. Independent testing put its accuracy on general-purpose content at 53 percent. The Bureau of Consumer Protection's director said the product "did no better than a coin toss." The order bars accuracy claims without competent and reliable evidence (FTC, April 2025).

Testimonials that look independent may not be. accessiBe paid $1 million to settle allegations that it overstated what its automated accessibility tool could do. The complaint also alleged it formatted third-party articles and reviews to look like independent opinions, without disclosing its connection to them (FTC, January 2025). If a consultant's case studies all live on sites you've never heard of, check who published them.

Review generation is its own category of trouble. Rytr sold a service that produced unlimited detailed consumer reviews from thin input. The FTC charged that this equipped subscribers to pollute the market with fake reviews. DoNotPay paid $193,000 over claims its service could substitute for a lawyer.

Enforcement did not stop with that sweep. A review published in August 2026 found the FTC had brought more than a dozen cases tied to AI washing in the preceding year. Attention has extended to business-to-business marketing claims (Holland & Knight, August 2026).

The practical test is short. Ask for the evidence behind any performance number in the pitch, and ask where the testimonials were published and who paid for them. A firm that can answer both in one email is probably fine. One that treats the questions as an insult has told you what you needed to know.

When Marketing Isn't Where the Money Leaks

Sometimes it is, and the data says so plainly. In the Federal Reserve's survey of 6,525 small employer firms, reaching customers and growing sales was the most commonly reported operational challenge. It ranked ahead of hiring or retaining qualified staff (Federal Reserve Banks, 2026). Demand is the number one problem small businesses name. Anyone who tells you marketing is a distraction for a business like yours is arguing with the best evidence available, and we won't do that.

So the question isn't whether marketing matters. It's whether the thing labeled "AI marketing" is aimed at your actual leak.

The Census working paper is unusually revealing here, because it measured two layers separately. At the firm level, the most common place AI gets deployed is Sales and Marketing, named by 52% of adopting firms. But at the worker level, the leading generative AI uses are writing, document analysis, and information search. The authors also found the two layers come apart. Worker task use sometimes happens with no formal firm-level adoption. And firm-level adoption sometimes happens with no worker task use (Census Bureau, April 2026).

Sit with that for a second. The budget line says marketing. The work being done is document and text handling. Which means a good part of what you buy as "AI marketing" is text automation pointed at the marketing function. Point that same capability at intake, quoting, and scheduling and it gets easier to measure, and closer to money already committed.

So run one diagnostic before you sign anything. Ask what share of last month's inquiries got a same-day response, and how many quotes went out without a follow-up. If you can't answer, that's the finding. Adding demand to an operation that drops inquiries pours more water into a leaking bucket, and it does it at a cost per lead.

If your follow-up is already tight and you're genuinely out of demand, marketing is the right target. Buy it with a baseline attached. If you don't know which situation you're in, that's what a readiness assessment is for. It costs far less than a year of the wrong retainer.

Eight Questions to Ask Before You Sign

Take these to the first call. The answers sort the field faster than any proposal review.

1. What will you measure before you change anything, and how? No baseline, no engagement. This is the one non-negotiable.
2. Which of these features does my current software already include? A consultant who has read your subscriptions before quoting is doing the job.
3. What is the recurring cost after you leave? Metered AI credits and add-on tiers are real line items. Get them in writing.
4. How many leads or conversions per month do I have, and is that enough to detect the lift you're promising? If they haven't done this arithmetic, they're guessing.
5. Who on my team owns this in month four? Name a person. If the answer is "we handle it," you're renting, not building.
6. What evidence supports the performance numbers in this pitch? The FTC standard is competent and reliable evidence. Yours can be too.
7. Which channel is this aimed at, and what are you assuming about AI search? A plan built on 2022 click-through assumptions is a plan built on sand.
8. What would make you tell me not to do this? A partner has an answer ready. A vendor doesn't have one at all.

For the broader version of this test across every kind of AI engagement, see what makes a good AI consulting partner versus a typical vendor.

When to Walk Away

Walk when the proposal has no baseline and no named metric. Walk when the deliverable is measured in posts per month. Walk when nobody on your side is assigned to own the result, because an engagement without an internal owner will quietly die in month four no matter what you paid for it. And walk when the scope duplicates features your subscriptions already include and nobody can say why theirs is better.

And walk when the pitch assumes marketing is your constraint without having looked at your operation. That one's worth spelling out. A firm that arrives with a marketing answer before asking a single question about how work moves through your business has diagnosed nothing. They've inventoried their own services.

Sometimes marketing genuinely is the constraint. Then the answer is yes, and the engagement should still come with a baseline.

Your Next Step

Do two things this week, in this order, and both are free.

Open your software subscriptions and list every AI feature you already pay for. Then count last month's inbound inquiries, and how many became jobs. That's your baseline and your volume. Those two numbers decide whether anyone — a consultant, an agency, or you — can ever prove the work helped.

Take both to the first call. If the consultant can work with them, you're talking to someone useful. If they'd rather steer back to capabilities and case studies, you've just saved yourself a year of retainer.

Not sure whether marketing is even the right target? Start with a readiness assessment, or work through which AI consulting company you should choose.

Frequently Asked Questions

Is AI marketing consulting worth it for a small business?

It's worth it when you're buying a measured baseline and a process your team keeps running after the invoice clears. It's rarely worth it when you're buying content volume. Your software likely generates content already, and the search channel that rewarded volume is sending fewer clicks than it did.

Don't I already have AI marketing tools?

Very likely. Mailchimp lists generative AI features as a no-additional-cost add-on on its $20-per-month Standard plan. Klaviyo includes an AI subject line generator on its free tier, and Jobber includes AI chat on all plans. Check your subscriptions before you scope anything.

How much does AI marketing consulting cost?

Clutch draws on a directory of more than 100,000 digital marketing companies. It reports US, Canadian, and Australian agencies billing $100 to $149 per hour, with a typical project landing between $10,000 and $49,999. Advertising media spend sits outside those fees and has to be budgeted separately.

Can I actually measure whether AI marketing worked?

Not as a conversion-rate test, at most small-business volumes. Detecting a 10% relative lift on a 10% baseline needs roughly 8,100 attempts per version by the NIST formula. Measure countable events instead: same-day response rate, hours to quote, follow-ups completed.

Should I fix marketing or operations first?

Find the leak first. Reaching customers is the most commonly named operational challenge in the Federal Reserve's small-business survey, so marketing is often right. But if inquiries are dying before anyone follows up, more inquiries cost money and change nothing.

Continue Learning

The buyer decision:

• Which AI consulting company should I choose?
• Is it worth hiring an AI consultant for a small business?
• Best AI consulting firms for small businesses
• What makes a good AI consulting partner vs. a typical vendor

Cost, readiness, and proof:

• How much do AI consultants charge?
• AI readiness assessment services
• How to choose an AI consulting service for business ROI
• Why do AI projects fail?

---

Sources

All sources retrieved 2026-08-19.

• Federal Reserve Banks, 2026 Report on Employer Firms: Findings from the 2025 Small Business Credit Survey, 25 March 2026 (fielded 3 September to 14 November 2025; 6,525 responses; nationwide convenience sample of firms with 1–499 employees), retrieved 2026-08-19 — https://www.fedsmallbusiness.org/reports/survey/2026/2026-report-on-employer-firms
• Kathryn Bonney, Cory Breaux, Emin Dinlersoz, Lucia Foster, John Haltiwanger and Keith Pande, The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks, NBER Working Paper 35141 and Census Center for Economic Studies CES-WP-26-25, April 2026, retrieved 2026-08-19 — https://www.nber.org/papers/w35141
• U.S. Census Bureau, Large Firms With at Least 20 Employees Biggest AI Users, Business Trends and Outlook Survey, 26 May 2026, retrieved 2026-08-19 — https://www.census.gov/library/stories/2026/05/ai-use-businesses.html
• Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, 22 July 2025 (900 U.S. adults on KnowledgePanel Digital; 68,879 searches observed March 2025), retrieved 2026-08-19 — https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
• Ryan Law and Xibeijia Guan, AI Overviews now correlates with a 58% lower average clickthrough rate, Ahrefs, 4 February 2026 (300,000 keywords; aggregated Google Search Console data; December 2023 versus December 2025), retrieved 2026-08-19 — https://ahrefs.com/blog/ai-overviews-reduce-clicks-update
• Gregory Druck, Jose Luis Paredes, Bevin Benson and Ethan Smith, More Articles Are Now Created by AI Than Humans, Graphite, October 2025 (43,000 URLs sampled from CommonCrawl), retrieved 2026-08-19 — https://graphite.io/five-percent/more-articles-are-now-created-by-ai-than-humans
• The CMO Survey, Firm and Industry Breakout Report, 35th edition, March 2026 (Christine Moorman, Duke University Fuqua School of Business; 308 U.S. marketing leaders, 97% VP-level or above; fielded 7–29 January 2026), retrieved 2026-08-19 — https://cmosurvey.org/results/
• Federal Trade Commission, FTC Announces Crackdown on Deceptive AI Claims and Schemes, 25 September 2024, retrieved 2026-08-19 — https://www.ftc.gov/news-events/news/press-releases/2024/09/ftc-announces-crackdown-deceptive-ai-claims-schemes
• Federal Trade Commission, FTC Order Requires Workado to Back Up Artificial Intelligence Detection Claims, April 2025, retrieved 2026-08-19 — https://www.ftc.gov/news-events/news/press-releases/2025/04/ftc-order-requires-workado-back-artificial-intelligence-detection-claims
• Federal Trade Commission, FTC Order Requires Online Marketer to Pay $1 Million for Deceptive Claims that its AI Product Could Make Websites Compliant with Accessibility Guidelines, January 2025, retrieved 2026-08-19 — https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-order-requires-online-marketer-pay-1-million-deceptive-claims-its-ai-product-could-make-websites
• Holland & Knight, "Operation AI Comply" 2 Years Later: Continued Enforcement Against Misleading Claims, 18 August 2026, retrieved 2026-08-19 — https://www.hklaw.com/en/insights/publications/2026/08/operation-ai-comply-2-years-later-continued-enforcement
• National Institute of Standards and Technology and SEMATECH, e-Handbook of Statistical Methods, section 7.2.4.2, "Sample sizes required" for testing proportions, retrieved 2026-08-19 — https://www.itl.nist.gov/div898/handbook/prc/section2/prc242.htm
• National Association of Home Builders, NAHB Census Reveals Small Businesses Continue to Drive Home Building, 26 August 2025 (member census; median builder member started six homes in 2024, median revenue $3.7 million, median six employees), retrieved 2026-08-19 — https://www.nahb.org/blog/2025/08/nahb-member-census-builders-in-2024
• National Association of REALTORS®, Experienced REALTORS® Anchor the Industry as Housing Affordability Remains Top Hurdle, 2026 Member Profile, 25 June 2026, retrieved 2026-08-19 — https://www.nar.realtor/newsroom/experienced-realtors-anchor-the-industry-as-housing-affordability-remains-top-hurdle-new-nar-report
• Rafael Damasceno, What to Do When There's Not Enough Traffic for A/B Testing, Seer Interactive, 25 March 2022, retrieved 2026-08-19 — https://www.seerinteractive.com/insights/not-enough-traffic-for-ab-testing
• Evan Miller, How Not To Run an A/B Test, 18 April 2010, retrieved 2026-08-19 — https://www.evanmiller.org/how-not-to-run-an-ab-test.html
• Clutch, Digital Marketing Agency Pricing Guide (based on a directory of more than 100,000 digital marketing firms plus first-party client reviews), retrieved 2026-08-19 — https://clutch.co/agencies/digital-marketing/pricing
• Intuit Mailchimp, Compare marketing plans, retrieved 2026-08-19 — https://mailchimp.com/pricing/marketing/compare-plans/
• HubSpot, Marketing Hub pricing, retrieved 2026-08-19 — https://www.hubspot.com/pricing/marketing
• Semrush, Pricing, retrieved 2026-08-19 — https://www.semrush.com/pricing/
• Jobber, Pricing, retrieved 2026-08-19 — https://www.getjobber.com/pricing/
• Klaviyo, Pricing, retrieved 2026-08-19 — https://www.klaviyo.com/pricing

On the two adoption figures. The 18% and the 46% come from different instruments and are not interchangeable. Census BTOS asks about firm-level deployment in a business function; the Federal Reserve's survey asks whether the business or its employees currently use AI. We've printed both and explained the gap rather than choosing the more dramatic one. The Federal Reserve sample is a convenience sample, not a probability sample, and the report says so.

On the marketing-spend table. These are means from cells of 19 to 30 firms with standard deviations close to the means, drawn from a sample of senior marketing leaders. A company with a VP of marketing is not a typical 30-person contractor. We've published the cell sizes and dispersion so the numbers can't be mistaken for a planning benchmark, and we recommend against using them as one. Figures circulating online attribute small-firm marketing budgets of 15.6% or 12.2% to this survey. They do not match the 2026 report. We read the primary and printed what it says.

On the click-through evidence. Pew's 8% versus 15% compares searches that did and did not surface an AI summary, and summary-triggering queries skew longer and more informational. It is a strong association, not a controlled experiment. Ahrefs describes its own result as a correlation. Google disputes the publisher traffic-loss framing as resting on incomplete data. We've reported the direction as well evidenced and the magnitude as contested.

On the sample-size table. Only the first row is NIST's published worked example. The other three rows are our own arithmetic on the handbook's formula for testing proportions, at the same one-sided 5% significance and 90% power. We reproduced NIST's published example exactly before computing the others. The formula's source is printed so the figures can be audited rather than taken on trust. They describe a two-arm test of a single conversion rate and are illustrative of the order of magnitude, not a substitute for a test plan.

On the volume anchors. NAHB's six housing starts and NAR's nine transaction sides are counts of completed work, not inquiries, and lead counts run higher than job counts. We use them because they're primary member-census figures and because no Tier 1 to 3 source publishes monthly lead volumes for these trades. The comparison is about orders of magnitude.

On the cost figures. The Clutch ranges come from a commercial directory. That's a weaker instrument than a survey with a disclosed sample frame, so we've named it rather than presenting a median as settled. We publish only the two figures the guide actually states — the hourly band and the project range — and no monthly retainer number, because the page does not give one. Several retainer medians circulating online are attributed to sources that do not contain them, and one widely cited agency benchmark report has no pricing data in it at all. We've excluded every one of them.

On vendor pricing. Feature inclusions and prices were read on each vendor's own pricing page on the retrieval date and change frequently. Verify against your own account before making a decision. Naming these products is not an endorsement; Avolis is stack-agnostic and we don't resell any of them.

On FTC matters. Case descriptions summarize allegations and settlement terms as published by the Commission. Settlements are not findings of liability.

On first-party claims. Statements describing what we see in our own diagnostics and client builds are Avolis's own observation, not independent research, and are not offered as benchmarks.

---

About Avolis Research Group

Avolis Research Group is Avolis's in-house research practice, focused on how operations-heavy small and mid-sized businesses actually adopt AI. It synthesizes primary economic research, government survey data, and results from real implementations into practical, vendor-neutral guidance.

More about Avolis and how we work · Get in touch