Skip to main content

CHOOSING & WORKING WITH AN AGENCY · September 2026 · ~11 min read

Red flags in an agency proposal

The warning signs in a proposal are all versions of one thing: a promise you cannot check. A promised first page ranking, deliverables with no unit attached, claims about a relationship with Google, and pricing that does not connect to a scope. Read the document looking for what you would be able to verify in month three.

A proposal is the only artifact you get before money changes hands. It is written carefully, by people who write them often, and it will be the clearest picture of how this firm thinks that you ever receive for free.

Read it the way you would read a vendor contract for your kitchen. Not for tone. For what it obligates them to actually do.

01

What is the single biggest warning sign?

A guarantee. Any guarantee, on rankings, on traffic, on leads.

The reasoning is not that guarantees are tacky. It is that the outcome depends on a system nobody controls, so the only way to make the promise safe is to define it down to something meaningless. Ranked first for a phrase nobody searches. Traffic that counts bots. There is a full explanation of why an SEO guarantee should make you more suspicious rather than less, and it is worth reading before you sign anything containing one.

Close behind it: any proposal that claims a special relationship with Google. Ad platform certifications exist and are checkable. Influence over organic ranking does not exist and cannot be sold. Google publishes the sentence for you in its own documentation on AI features, warning readers to be wary of third party tools that promise ranking success or claim to use internal Google metrics, because no third party tool has access to its internal ranking or AI systems.

02

Which line items cannot be verified?

Go through the scope with a pen and mark every item that has no unit.

"Ongoing optimization." "Continuous monitoring." "Technical improvements as needed." None of those can be counted at the end of a month. Compare them to items that can: four pages published, twelve directory listings corrected, review requests configured on the booking system, three schema types added to five templates.

A scope you cannot audit is a scope that shrinks quietly. Not always because anyone intends it, but because unmeasured work is the first thing to lose when the team gets busy.

Ask for a unit on every line. Content becomes a number of pieces, with topics and an approval step. Reporting becomes a named set of metrics. If a line resists having a number attached, ask what proof of completion looks like. A reasonable agency will answer this in one email, and the answer usually arrives shorter than the original scope once the padding comes out.

03

How do I price the page speed line?

Worked example

Almost every proposal has one, it is almost never attached to a measurement, and it is the cleanest line to work through because the public evidence is unusually good.

Step one, get the current number before you agree to a target. Core Web Vitals thresholds are published. Largest Contentful Paint is good at 2.5 seconds or under and poor above 4 seconds, Interaction to Next Paint is good at 200 milliseconds or under, and Cumulative Layout Shift is good at 0.1 or under. All three are assessed at the 75th percentile of real visits, not at a lab score from a testing tool. Ask for your field number at that percentile. If the proposal cannot produce it, the line is not priced against anything.

Step two, size the gap. Say your Largest Contentful Paint sits at 4.1 seconds and the proposal targets 2.4. That is a gap of 1.7 seconds, and it is a real gap, since you are currently in the poor band and the target is inside the good band.

Step three, resist the multiplication, and here is where most pitches go wrong. Bing ran controlled slowdown experiments, deliberately delaying the server for a share of users, and measured that every 100 millisecond speedup improved revenue by 0.6%, that a 250 millisecond delay cost about 1.5%, and that 500 milliseconds cost about 3%. It is tempting to take your 1.7 seconds, divide into seventeen increments, multiply by 0.6% and arrive at 10.2%derived.

Do not do that, and do not accept a proposal that does. The measurements were taken on a search engine at web scale, over a small range near an already fast baseline, on desktop, with the paper's authors stating plainly that mobile is a different problem. Extrapolating a linear rate across 1.7 seconds on a local business site is arithmetic wearing the costume of evidence.

Step four, ask where on the page the delay lives. In the same body of work, Bing delayed elements in the right hand pane by 250 milliseconds across almost 20 million users and could not detect an impact on key metrics. Position decided whether the milliseconds were worth anything. A speed line that does not say which elements it is fixing is not a plan.

Google's own study with Deloitte and Fifty Five is the best available number on the other side, and it needs its date attached. Across 37 brands in retail, travel, luxury and lead generation, measured hourly over four weeks in October and November 2019, a 0.1 second improvement in mobile load time was associated with retail conversions higher by 8.4% and average order value higher by 9.2%. That is Google measuring a case for speed, on large brands, seven years ago. Useful, and not a promise about your site.

04

Which speed claim in the deck is simply false?

The famous one. Google showed thirty results instead of ten, traffic dropped twenty percent, and it was because the page was half a second slower.

It appears in nearly every page speed pitch and it has been in circulation for years. The trouble is the causal half. The same Microsoft, Amazon and LinkedIn authors who ran the controlled slowdown tests addressed it directly: at Bing, half a second of delay was worth about 3% of revenue, not twenty. They state that they suspect the speed factor accounts for only a small percentage of the loss.

Which means the missing seventeen points came from showing people thirty results, not from the delay. The claim is real research, retold with the wrong cause attached, and the wrong cause is the one that sells a speed project.

The same paper documents the mirror image at Bing. Truncating the results page to four algorithmic results for one class of query, across more than eight million users, moved abandonment not at all. Extending it to twenty results, across more than three million users, reduced revenue by 1.8% and cut pagination by 18% while abandonment again did not move. Page length was not the thing. What the length cost was.

05

What should the pricing section tell me?

That the fee moves when the scope moves, and not otherwise.

You are not looking for a low number. You are looking for a clear relationship between what you pay and what gets produced. Three things to check.

First, whether onboarding and ongoing work are separated. A lot of the value in the first sixty days is one time work. If it is bundled into a monthly fee, you are paying for setup for as long as you stay.

Second, whether media spend is separate from management fees. Ad budget passing through as revenue makes the invoice look bigger and makes the management fee harder to evaluate.

Third, whether the proposal says what happens when you want something extra. Silence there becomes an awkward conversation in month two.

What you cannot do is compare the fee to an industry benchmark, because there is not one. No survey of agency management fees or website build pricing with a disclosed sample and method exists. Every range in circulation traces to agency pricing pages citing other agency pricing pages. Judge the fee against the scope in front of you.

06

Which claims should I check before signing?

Three specific ones, all checkable in under an hour.

Proprietary tools and algorithms. Ask what the tool does that a standard rank tracker does not. Sometimes the answer is genuinely good. Sometimes the tool is a reporting dashboard with a brand name on it, which is fine, but it is not a technical advantage.

Location coverage. If a proposal offers to build profiles or listings in cities where you have no staffed address, stop. Whitespark's 2026 expert survey added several practices to its suspension risk list, including showing an address on a profile when the business is a service area business, setting service areas larger than about two hours of driving, and overlapping service areas across multiple profiles. That is a policy violation with real consequences for your existing profile, and the reason fake listings keep appearing in the map is worth understanding before you accidentally commission one.

Anything sold as a technical requirement for AI search. Google's documentation states that a page needs to be indexed and eligible to be shown with a snippet, with no additional technical requirements, and says you do not need new machine readable files, AI text files, markup or Markdown, because Search ignores them. If a proposal charges for AI specific markup or files, compare it against what Google actually publishes about AI search before you pay for it.

07

What about the terms attached to the back?

The commercial pages are usually where the risk sits, and they are usually the pages nobody reads.

Look for the initial term, the notice period, and whether it renews automatically. Look for who owns the website, the content, the ad accounts, and the analytics property if you leave. Look for whether the agency can subcontract without telling you.

None of these are automatically bad. A six month initial term is defensible for work that takes months to show anything. What is not defensible is a twelve month lock with a sixty day notice window buried in a paragraph, on a scope you cannot audit. The specific clauses worth negotiating, and the ones not worth the fight, are a short list.

08

What to do this week

Print the proposal. Three passes with a pen.

First pass, circle every deliverable with no number attached. Send them back as one email asking for a unit on each.

Second pass, underline every claim about outcomes. Ask what happens if it does not occur, and whether anything in the document changes as a result.

Third pass, read only the terms at the back. Write down the notice date you would need to hit to leave at month seven. If you cannot work it out from the document, that is your answer about how the document was written.

Then send the one line that any agency writing a speed or performance line should be able to answer, and that includes us. What is my current Largest Contentful Paint at the 75th percentile from field data, and what will it be when you are done? SEOD writes proposals with units on every line and would still have to produce that number before quoting the fix. An agency that quotes a speed project without the field measurement is selling a project, not an improvement.

If you have been through a bad engagement before, your instinct will be to over index on the things that went wrong last time. That is understandable and it is also how people miss new problems, which is why rebuilding trust after a bad agency experience needs its own approach.

Be honest with yourself

When you do not need this

If the proposal is for a fixed scope project under a few weeks, most of this does not apply. Check the deliverable list and the ownership clause, agree, and move on.

If you have worked with this firm before and they delivered, a light read is enough. Trust earned through a completed project is better evidence than any document.

And if the proposal is genuinely good and you are still hunting for something wrong with it, notice that. Some agencies write clear scopes and honest terms. Finding one and then talking yourself out of it costs you a quarter.

Sources

Related reading

12

Questions about a proposal sitting on your desk?

Email me at eric@seod.com with the proposal, pricing redacted if you prefer, and I will mark which lines are real deliverables and which ones cannot be checked at the end of a month. I will not bid against whoever wrote it, and I will not ask for the introduction. If it is a good proposal, my note will be short and will say so.

There is more on hiring and working with agencies if you are earlier in the process.

Call Eric Email Eric