The guru discount: how to read marketing advice without getting burned

Most marketing advice is published by someone who gets paid in the thing they are recommending.

Plenty of it is good. It also arrives with a direction built in. A survey written by the company being surveyed arrives the same way. What you are reading is a finding that somebody chose to publish.

Other fields have a procedure for reading evidence that leans. Doctors apply one to drug studies. Journalists apply one to press releases. Marketing rarely applies one. So the field keeps repeating numbers whose source nobody has checked.

I call the marketing version the guru discount. It takes about 10 minutes per claim. It is learnable in an afternoon. It is the habit that most reliably separates owners who spend well from owners who fund other people's experiments.

The person who taught you the tactic is usually paid in that tactic

Start with the oldest rule in the field. A prospect needs to see your message seven times before they act.

Search for where that number comes from. Page after page traces it to movie studios in the 1930s, counting how often audiences saw a poster. None of those pages names the studio, the study, the sample, or the measurement. The citation is other pages that also do not name it.

Now ask who benefits from the rule staying alive. Ad platforms sell exposures. Agencies bill for campaigns measured in touches. Retargeting vendors sell the seventh touch specifically. The one party with an interest in fewer, better touches is the person paying for them. That person is rarely in the room when the rule gets taught.

A claim that survives on repetition rather than measurement is usually one that pays somebody to repeat it. No one has to lie for that to happen. An unmeasured number stayed in circulation because circulating it was profitable, and checking it was nobody's job.

The same pressure shapes advice that is not folklore. Which tactics get written up. Which results get highlighted. Which failure gets left out of the case study, and which caveat gets cut for length. Each of those choices bends toward the publisher's business. Scored against practitioner ledgers, a short list of tactics compounds. Most of the rest breaks even.

Telling you about the bias does not remove the bias

The usual answer to conflicted advice is disclosure. Label the sponsored post. Say you sell the software. Put the affiliate note at the top, and the reader can adjust from there.

Readers do not adjust enough. Daylian Cain, George Loewenstein and Don Moore ran the experiment, and disclosure left advice recipients worse off. Two things happened at once. People told about an advisor's conflict discounted the advice a little, and nowhere near enough. The advisors, meanwhile, saw their conflict acknowledged and felt licensed to push harder. The advice itself got more biased.

Look at the shape of that. The warning label made the product stronger and the buyer more comfortable.

The same researchers found one thing that worked. When people saw the conflicted advice set directly beside advice from someone with no stake, they corrected properly. The comparison did what the disclaimer could not.

That finding is the whole method. Knowing a claim is conflicted does not fix the claim. You fix it by putting an unconflicted account next to it before you decide.

The guru discount is four questions

Run the guru discount on any piece of marketing advice before you act on it. The four questions take longer to read than to use.

1. Who gets paid if I believe this, and in what unit? Impressions, seats in a cohort, a monthly retainer, software licenses. The unit tells you which way the advice will lean before you read a word of it.
2. What is the original source, and does it say what the claim says? Trace the number back until you reach the organization that collected it. If the trail ends at another blog post, you have folklore.
3. Who is running this with a business like mine, and what do they report? Not the testimonials on the sales page. The people discussing it where nothing is being sold.
4. What would have to be true for this to work here? Most tactics carry unstated conditions. An existing list. A budget floor. A category where buyers search, or a founder willing to be on camera.

The fourth question is where most of the money gets saved. A tactic can be effective and still be unavailable to you. The person it worked for may have started with an audience, a war chest, or a decade of relationships. I have changed my mind about tactics I once recommended more than once. A hidden starting advantage was usually the reason.

The score most firms optimize was built by the companies selling the software

Take a live claim, the one you will meet in your first conversation with almost any search agency. Your domain authority is too low, and raising it is the work.

Question one. Domain Authority is a Moz product. Ahrefs sells Domain Rating. Semrush sells Authority Score. Each is a number produced by a company that sells a subscription to see the number.

Question two. Trace it to the organization whose behavior it claims to predict. Google's search representatives have said for years that it uses no third-party authority score in crawling, indexing, or ranking. Moz's own documentation describes the score as a comparative tool rather than an absolute one.

Question three. Practitioners who work on client sites will tell you the score tracks something real. It measures link strength, and link strength does correlate with visibility. They will also tell you what happens to firms that chase the number directly. Bought links, inflated scores, and no movement in the thing the owner wanted.

Question four. For the claim to earn a budget line, raising a vendor's proxy would have to raise your rankings. On its own, it does not. The proxy moves when the underlying work happens, which means the underlying work was the job all along.

Notice what the discount did here. It left the tools standing, because they are useful and I use them daily. It identified a measuring instrument that got sold to the field as a goal. A plan built on raising that score is a plan built on somebody else's dashboard.

Practitioner reports are the best correction available, and they skew too

The third question sends you to forums, groups, and communities. That is where people describe results with nothing riding on your belief. It is the closest thing marketing has to the unconflicted second opinion the researchers found so effective.

It carries a distortion of its own, and you need the shape of it. Online reports cluster at the extremes, a pattern documented across platforms and categories as a J-shaped distribution. The person who had a spectacular result posts. The person who got burned posts. The person for whom it worked moderately well, at a fair price, says nothing.

So read forums the way you would read a jury pool. Weight the long, boring, specific posts, the ones naming their spend, their category, their timeline, and their own mistakes. Discount the ecstatic and the furious equally. Check whether the account has a history in the field or arrived last week. When a claim is disputed, find the reply that changes the original poster's mind. That reply is usually where the condition hiding inside the tactic gets named.

Two searches do most of the work: the tactic plus "not working," and the tactic plus the current year. Both surface the people the marketing web has no reason to quote.

What a chatbot changed about checking a claim

Tracing a statistic to its source used to take an evening in a library, then an hour of searching. It now takes a few minutes, which changes the economics of the guru discount completely. Ask the assistant for the original study behind a claim. Then open the link it hands you and read the date and the sample size yourself.

Two cautions, both load bearing.

Check the link before you trust the answer. The European Broadcasting Union and the BBC had journalists evaluate more than 3,000 assistant answers about news. In that study, 31% had significant sourcing problems, meaning attribution that was missing, misleading, or wrong. A citation can look right and still fail when you click it.

Then watch your phrasing, because your own opinion leaks into the answer. Researchers tested seven open language models by adding a statement of belief to a question. The models then agreed with an incorrect answer 64% of the time, ranging from 47% to 95%. Those models were smaller than the assistant you are likely using. Read the figure as a warning about phrasing rather than a measurement of your tool. The instruction it supports is simple enough. Skip "LinkedIn ads are worth it for a firm like mine, right?" Ask what the evidence says for and against LinkedIn ads for a small professional firm. Then ask for the strongest case against it.

One move that does not count as checking: asking the assistant whether it is sure. That returns another answer from the same system, and agreement is not verification.

What the guru discount will not do

The guru discount will slow you down. Ten minutes a claim adds up. Some weeks the checking will feel like a way of avoiding the work.

It will sometimes cost you. Applied strictly, it will occasionally talk you out of something that would have worked for you. The evidence was thin at the moment you looked, and the tactic was early. That is the price of the method. It is worth paying, because the errors it prevents are the expensive kind.

It cannot tell you whether a tactic fits your business. It tells you whether the evidence behind a claim is as strong as the claim sounds. Fit is a separate question, and it stays yours to answer.

And it will not make you right. It makes you wrong less often, and less expensively. Over a decade, that separates a budget that compounds from one that funds other people's old experiments.

Two failures sit close to this one, and the discount alone will not catch either. Advice can be accurate when written and expired by the time you read it. A tactic's edge is how few people are using it. And a case study can be true in every particular while proving only that something is possible. How often it happens is a different claim.

Start with me

I sell marketing services. The argument above says you should discount people who sell marketing services.

So run the four questions on it. Who gets paid if you believe this, and in what unit. Whether the studies say what I said they say, which the links make a five-minute job. Who else works this way, and what their clients report. And what would have to be true for the method to help a business like yours.

An answer that holds up teaches you the procedure. An answer that does not teaches you something more useful than anything I could have told you.