How to Verify an SEO Agency's Results Before You Sign

Ask for the full time series, not the best week: a screenshot of one ranking on one day is the single easiest artifact in this industry to manufacture, and a Search Console export covering every week of an engagement is the hardest. Everything else on this page is a variation on that one request.

The reason it matters more than it used to is that the metric most agencies report has quietly stopped predicting revenue. SparkToro’s 2026 analysis found that fewer than a third of Google searches still send a click to the open web, so an agency can move you up a ranked list and deliver nothing you can bank. When the headline number no longer maps to outcomes, the burden shifts to evidence you can inspect yourself.

What evidence should an SEO agency be able to produce?

A defensible answer has four parts: a continuous data series rather than a snapshot, access to the source system that produced it, a named change with a date attached, and the periods where the number went the wrong way. An agency that can supply all four is describing work it actually did. One that supplies only the first is describing a chart.

The continuity requirement does most of the work. Search Console keeps a rolling sixteen months of query and page data, and its Performance report is exportable in full, as Google’s own documentation sets out. That means “here is every 28-day window since we started” is always an answerable request for genuine work — the data exists whether or not the agency likes what it shows. When the reply is a cropped image instead, the constraint is rarely technical.

The dated-change requirement is the one buyers skip. A position that improved tells you nothing about causation on its own, because SERPs drift, competitors fall out, and seasonal demand moves whole query families at once. What makes a result legible is a specific intervention on a specific date with the series either side of it. Our own clearest example: we rewrote one headline on our Salt Lake City page on 2026-08-03 and changed nothing else — no links, no new content — and the page moved from average position 78.2 to 33.0 across the four 28-day windows that followed.

Which numbers are easy to fake, and which are hard?

Rankings screenshots, “impressions” with no denominator, and traffic charts with the axis cropped are trivially gameable; Search Console exports, Analytics conversion counts tied to a tracked destination, and dated before-and-after series are not. The distinction is whether the number comes from a system the agency controls or one you can log into.

Artifact	How hard to manufacture	What to ask instead
Ranking screenshot	Trivial — personalized, localized, one moment in time	The 28-day average position series from Search Console
”Impressions up 300%“	Easy — a low base makes any number look large	Absolute impressions, both periods, and the query list
Traffic graph	Easy — cropped axes, undated ranges, bot traffic included	Sessions by landing page with engagement rate beside them
Search Console export	Hard — Google’s own data, timestamped, complete	Read-only access to the property for the engagement
Conversions to a tracked path	Hard — requires an actual booking or form event	Key-event counts by landing page in GA4

The middle row is worth dwelling on, because it is where honest agencies accidentally mislead. Traffic is not the same as human traffic. On our own site, most sessions in any given month arrive direct with a bounce rate near 1.0 and an average duration of roughly zero seconds, which is a signature of automated traffic rather than buyers. The eight sessions a month that arrive from Google organic run about 109 seconds and produce the conversions. An agency reporting the big number without the quality split is not lying, but it is not telling you anything either.

What should you ask about AI visibility specifically?

Ask which prompts the agency has logged, how often, and what the answer format was — then ask to see a run where the client was absent. AI visibility is the newest claim in this market and the least standardized, which makes it the easiest place to sell a screenshot.

There is no ranked list behind an AI answer, so “we got you to number one in ChatGPT” is not a coherent statement. What is coherent is a sampling log: the same buyer question run on a fixed schedule, with the businesses named, the sources cited, and the answer format recorded each time. Format matters more than most buyers expect. In our own weekly runs, the assistant sometimes answers a local hiring question with prose and sometimes with a review-ranked business panel, and those two formats are won by completely different work — one by content and citations, the other by your Google Business Profile and your customers’ reviews.

Volatility is the other reason a single run proves little. Business Insider reported in August 2026 that Reddit citations in ChatGPT answers fell by roughly 86% within weeks, which is the kind of swing that can make any agency look brilliant or useless depending on the week they screenshot. We publish our own record on that basis: across ten weekly samples of the same Utah buyer prompt, ChatGPT named Flownomic in four and omitted us in six. The six are the informative half.

How do you tell a real explanation from a plausible one?

Ask why the number moved, and listen for whether the answer names a mechanism you could have checked yourself. “We improved your topical authority” is unfalsifiable. “Your service page’s headline said one thing and Search Console showed Google surfacing it for a different phrase, so we matched them” is a claim you can verify in ten minutes.

The research base here is thin but real, and an agency that has read it will tell you where it stops. Aggarwal et al. (Princeton, 2023) found that adding citations, quotations, and statistics to a page raised its visibility inside generated answers by up to 40% — a genuine, measured effect on the content side, and one that says nothing about local business panels. Separately, research from Trustpilot reported by TechRadar Pro in 2026 found businesses with 80 or more reviews appeared in over 75% of AI answers while those with none appeared in roughly 1%. Those two findings point at different levers, and a diagnosis that does not say which one applies to you is a template.

Watch for the reverse tell too. An agency that claims it can place you in review-ranked results without reviews is either misunderstanding the mechanism or proposing something that violates Google’s review content policy. Both are disqualifying, and the second is the more expensive mistake to inherit.

What does a good answer to “show me your results” sound like?

It sounds specific, bounded, and slightly disappointing. Real search work produces movement in stages and admits what it has not yet achieved, because the honest ceiling on most on-page interventions is a position, not a customer.

Here is our own version, stated the way we would want to hear it. One headline change moved a page 45 positions in four weeks, which demonstrates the diagnostic works. Those positions are in the twenties and thirties, which is page three, and page three does not produce clicks — closing that gap requires off-site authority, which is slower and less certain than anything on the page. We ran the identical change on a second city page and it reached position 29.6 before drifting back to 35.4 as the query family softened. We publish that drift because an agency that only shows you the up-weeks has told you nothing about the down ones.

The tell you are looking for is whether the agency volunteers a limit before you ask for one. When we sampled ChatGPT on a Utah hiring question this month, the model named us and then advised the buyer to demand hard evidence of local-business results rather than AI visibility screenshots. That is a reasonable thing for a buyer to want, and the agencies worth hiring will hand it over without being pressed. If you want that conversation about your own site, we do it in thirty minutes, free, and you keep the findings either way. Our own full series is published on our results page, and if you are shopping locally, our guide to choosing a Utah SEO company covers the market-specific version of these checks.

What we’d actually do

Ask for read-only Search Console access on day one, before you sign anything. It costs the agency nothing, it is reversible, and the reaction tells you more than the pitch deck did. Any agency doing real work will grant it in a minute, because the data supports them; any agency that hesitates is telling you the number they showed you does not survive contact with the source.

Then pick one query you actually care about and ask three questions: what position is it at today, what specifically will you change, and when will we look again. Write the answers down. In ninety days you will either have a series that moved after a named change, or you will have a vendor who has been reporting on impressions. That distinction is worth more than any case study either of you could produce, and unlike a case study, it is about your business.