The SEO Statistics Everyone Quotes, Traced Back to Source

I read a lot of SEO marketing. It comes with running a technical SEO business: I want to know what is being said to the people I am also talking to. And the same handful of statistics turn up everywhere, in agency decks, on pricing pages, in cold emails, always without a source you can actually follow.

So I spent a while tracing them. Not all of them survived.

This is not a piece about other agencies being dishonest. Most people repeating these numbers believe them, because they read them somewhere that looked credible, which had read them somewhere that looked credible. That is how a statistic gets laundered. But if you are about to spend money on the strength of one, you should know which ones stand up.

"76% of people who search locally on their phone visit a business within 24 hours"

The most quoted statistic in local SEO, in thousands of posts, most of them dated this year.

It comes from a study by comScore, Neustar Localeze and 15 Miles, published in 2014. Google restated it in a Think with Google piece in 2016, and almost every current citation traces back to Google quoting comScore rather than to the research itself. The original report is no longer easy to find online at all.

It is also US data, about mobile local search behaviour in general, not about any kind of business in Britain.

Real study, twelve years old, wrong country, and nobody quoting it has read it. If you see it in a pitch dated 2026, you have learned something useful about how carefully the rest of the pitch was assembled.

"SEO delivers an average 748% ROI"

Usually credited to BrightLocal, often paired with "87% report measurable revenue growth".

I looked at BrightLocal's statistics page directly. Neither figure appears on it. BrightLocal publish plenty of genuine research, including their Local Consumer Review Survey, which has run for years and is properly documented. This is not part of it.

The 748% figure is real, though, and it does have a home. It traces to First Page Sage, an SEO agency, describing the median return across its own clients' campaigns over a 36 month period.

That is a completely different animal from an independent benchmark. It is one agency's own book of business, self-reported, measured by the people being paid. It may well be accurate for those clients. It tells you nothing about what you should expect, and it is not research.

Somewhere along the way the attribution drifted from First Page Sage to BrightLocal, and it now circulates as though an independent research firm had established it across the market. If you see it credited to BrightLocal, the person quoting it has not checked.

"68% of local searches trigger an AI Overview"

The newest one, spreading fastest, and the most interesting to take apart.

It comes from Whitespark, May 2025, who ran 540 queries across six industries in three US cities: Houston, Phoenix and Denver. Real research, honestly reported.

Here is the part that gets left out. Whitespark broke their results down by what kind of query it was, and the categories behave completely differently.

The 68% is the blend of all three. On the queries that actually matter to a local business, someone looking for a supplier near them, an AI Overview appeared 15% of the time. A local pack appeared 93% of the time.

So the headline is not merely imprecise, it is close to backwards. Local intent queries were the least likely category in the study to show an AI Overview, and it is Whitespark's own data, in the same report, that says so.

UK data agrees. A June 2026 Ahrefs pull of 979 UK local intent keywords across 11 sectors:

Source: Studio36 Digital, UK Local SEO Statistics 2026. Note the small samples on legal and accountancy, 20 and 28 keywords, so treat those two as indicative.

If you are a roofer, an AI Overview appeared on none of the searches sampled. If you are a dentist, one in five. Those are two completely different situations, and one blended number hides both.

It also tells you where to spend. The local pack, meaning the map results, appeared on 92% to 97% of searches for every trade in that sample. For a plumber, your Google Business Profile is doing more work than anything else on this list. Anyone selling you AI visibility as your most urgent problem, while your map listing sits half filled in, has the order wrong.

"Near me searches have grown 500%"

Also appears as 150%, 200% and 900%, depending on the blog.

All of it originates from Google's Think with Google pages, and the underlying comparison period is July to December 2015 against July to December 2017. US data, about a narrow subset of "near me to buy" mobile searches. That is a ten year old comparison being recycled into posts dated 2026 as though it described something happening now.

The growth was real. The number is a fossil, and it has nothing to do with AI search.

"44% of local searchers click the map pack, 29% click organic"

I could not trace this to a named study at all. It appears across dozens of SEO sites in near identical wording, with a citation chain that ends nowhere.

The underlying idea is almost certainly directionally right, and the UK table above supports it. But the specific split is a number without a parent.

Every voice search statistic

This one has the best story behind it, and it is worth telling properly.

The famous claim is that half of all searches would be voice by 2020, usually credited to comScore. comScore never published it.

The actual lineage: in September 2014, Andrew Ng, then chief scientist at Baidu, told Fast Company he expected half of searches to be image or speech within five years. Not voice, and not comScore. Mary Meeker's 2016 Internet Trends report popularised "by 2020" as the date. Then a Campaign UK article in April 2016 attributed the claim to comScore with no source link, and it has propagated on that misattribution ever since.

A prediction by one person about image and speech became a comScore statistic about voice. Every current voice search figure I checked, "71% prefer voice to typing", "58% use voice weekly for local business", "76% of voice searches are local", leads to a roundup citing another roundup.

The keyword volume problem

A small one that tells you a lot. Looking for UK search volume for "accountant near me", I found two agency blogs. One said 3,500 a month. The other said 40,000 to 60,000 across local variations.

Neither showed a methodology. They disagree by more than ten times.

Volume figures in marketing material are tool estimates, and different tools disagree substantially. That is fine, as long as nobody presents one as a fact. If an agency shows you a volume number, ask which tool, which date, and whether it was UK targeted.

And llms.txt, while we are here

Not a statistic, but the same problem. llms.txt is a file you can add to your site to tell AI systems what it contains, and there has been a lot of enthusiasm about it.

Ahrefs monitored 137,000 domains and found adoption growing fast, up nearly nine times in a year to around 36,000 sites. They also found that 97% of published llms.txt files were never requested by anything at all. Of the 3% that were fetched, the biggest named fetcher after GPTBot was Anthropic's Claude Code, which is a developer tool, not an answer engine.

SE Ranking looked at roughly 300,000 domains separately and found no relationship with AI citations. Removing llms.txt from their model actually made it more accurate, meaning it was noise.

Google's John Mueller compared it to the keywords meta tag, the long abandoned tag that search engines stopped using decades ago because it was trivially gamed. Gary Illyes has said Google does not support it and has no plans to.

I still add one for clients who want it, because it costs nothing and might matter later. I would not sell it as a reason to hire anybody today.

The numbers I do use

Plenty of it holds up. These are sourced and dated, and I have said where they are not British, because most of the good research is not.

• Ofcom, Adults' Media Use and Attitudes 2026, published 2 April 2026, 7,533 UK adults: 54% of UK adults now use AI tools such as ChatGPT, Copilot or Gemini. It rises to 79% of 16 to 24 year olds. Proper sample, proper methodology, British.
• The CMA's Final Decision on Google's search market status, 10 October 2025, built on compelled disclosures from Google, OpenAI, Perplexity, Meta and Anthropic. As reported, AI assistant queries rose from 0 to 5% of Google Search volume in December 2024 to 10 to 15% by June 2025, while search grounded AI queries specifically remained at 0 to 5%. It also states that AI Overviews already appear on more queries than ChatGPT receives in total across every use case.
• That last point reframes the whole thing. The shift is Google answering more searches inside Google, not everyone leaving for ChatGPT.
• Pew Research Center, 22 July 2025. US data, and worth flagging because a lot of people quote it as though it were universal. Real browsing data from 900 US adults across 68,879 searches: users clicked a traditional result 8% of the time when an AI summary was present, against 15% when it was not. There is no UK equivalent study. If someone shows you a UK click loss figure, ask where it came from, because at least one widely circulated version is German data from SISTRIX relabelled as British.
• Ahrefs, 11 May 2026, tracked 1,885 pages that added JSON-LD structured data against 4,000 matched control pages, and found the change in AI citations was close to nil. Worth reading carefully, because the useful lesson is not the one people take from it. They tested one change on its own, on pages that were already heavily cited. What it shows is that no single tactic moves the needle by itself.
• What does correlate with being cited by AI, per an Ahrefs study of 75,000 brands: YouTube mentions at 0.737, branded web mentions at 0.656 to 0.709, branded anchor text, branded search volume. Domain Rating around 0.3. Raw backlink counts, weak. Correlation is not causation, but the shape is unmistakable: it is your whole footprint that decides this, not any one file on your server.

Those two findings sit together, and together they say something worth acting on. Being found now depends on the full picture: a site that engines can crawl and read, facts that agree with each other everywhere they appear, content shaped like the answers people actually ask for, and a business that gets mentioned in the places these systems draw from. Structured data is part of that. So is crawler access, so is site speed, so is what your listings say, so is whether anyone else on the web mentions you at all.

Which is why I do not sell any of it as a single item. There is no schema button, no AEO button, no GEO button. There is one job, making a business legible and correct to every system that decides who gets recommended, and it is done across all of those layers at once or it is not really done.

How to check a statistic in about two minutes

You do not need to be technical. Four questions:

• Who collected it? A named organisation, not "studies show".
• When? If nobody will tell you the year, assume it is old.
• Where? A great deal of UK marketing runs on US or German data. Consumer behaviour differs and so does the search market.
• How many? 540 queries in three cities is a real study with real limits. 7,533 respondents is something else. Both are useful, neither is interchangeable with the other.

Then click the link. Then click the link on that page. If you end up on a blog quoting a blog, you have found the bottom, and there is nothing underneath it.

What I would do about it, and what it costs

Here is the uncomfortable part, and it is the reason any of this matters. When a business is hard for search engines and AI systems to read, nothing dramatic happens. No error message, no drop you can point at in a report. People search, they get shown somebody else, and you never hear about it. That is exactly why it goes unfixed for years, and why it is worth dealing with properly rather than buying a package sold on a statistic that falls apart when you check it.

I handle the whole technical side of being found as one job: SEO, AI search, answer engine and generative engine optimisation together. Site structure and crawlability, structured data, metadata, canonicals, sitemaps and AI crawler access, all set correctly and then kept right every month as your site changes and the standards move.

It is £100 a month. First month free, no contract, and you deal with me directly rather than an account manager. Start here, or email me your website address and I will tell you exactly what is holding it back.