Read enough search ranking documents leak SEO threads this week and you'll come away with a list of tricks. Tweak this attribute, chase that score, add author boxes everywhere. That's the wrong lesson for an agency with small business clients. What the leaked documentation really describes is a system that pays attention to whether people choose a site, trust it across the board and come back to it. You can't fake that with on-page edits.
So here's the argument of this piece. The leak gives you a reason to stop selling isolated keyword wins and start selling demand. That means more people searching for your client by name, landing on pages worth staying on, and returning. Email and distributed content build that demand directly. Title tag rewrites don't.
Below you'll get what the documents show and where their limits are, what click signals and site-level authority mean in practice, and why brand demand is the one signal a florist, a dental practice or a local accountant can grow without outspending anyone. The second half covers how email feeds it, how to measure it and how to rewrite the retainer around it.
The leak confirms what clicks and brands were already doing, so stop selling keyword wins on their own
The search ranking documents leak SEO teams are picking through is internal API documentation from the dominant search engine, roughly 2,500 pages describing more than 14,000 attributes, which surfaced publicly earlier this week. It reads like a parts catalogue. It names fields the systems store about pages, sites, links, clicks and authors. It doesn't say how much any of them count. There are no weights in it, no ranking formula and no order of importance. Treat it as a list of what gets recorded, and the useful reading is simple. Click behaviour, site-wide reputation and recognition of the business as an entity all appear in the record, and every one of them favours a business people already know and look for. That's the case for selling demand, with keyword positions as one result of it rather than the product.
What the documents show, and what they can't tell you yet
Five things keep getting quoted in client emails and agency forums, and they're worth knowing precisely so you can talk about them without inflating them.
- A click-based re-ranking system, named NavBoost in the documents, with attributes such as goodClicks, badClicks and lastLongestClicks. Its existence wasn't new. It was described in public court testimony last year.
- A siteAuthority attribute, which points to some measure of quality or authority held at the level of the whole site, not only the page.
- Attributes that reference data from the company's own web browser, suggesting browsing behaviour is recorded in some form.
- Entity and author attributes, including fields for whether a page names an author and how entities on a page are identified.
- Demotion flags, covering things like exact-match domains, poor product reviews, anchor text that doesn't match the target page, and navigational signals.
Now the limits, which matter as much as the list. Nothing in the documentation shows the weight of any attribute. Nothing proves which attributes are live in ranking today rather than stored, tested or retired. Parts of it are likely out of date, since documentation of this kind lags behind the code. And the official response so far has been a caution that the material lacks context and may be outdated or incomplete, with no comment on any specific attribute. Everything beyond the field names is interpretation, including ours.
Why 'it's in the docs' does not mean 'it moves rankings'
An attribute existing tells you the system stores a value. It doesn't tell you the value changes where a page appears. A field might feed one small classifier, might only be used for a test, or might sit there unused because someone added it years ago and nobody removed it. Think of a customer database with a column for fax numbers. The column exists. It doesn't decide who gets a call back.
This matters for your contracts. If you tell a dental practice that you'll optimise for siteAuthority or that author boxes will lift rankings because the leak says so, you're promising something nobody outside the company can measure. When the result doesn't come, the client remembers the promise. Talk about the direction the documents point in, not about specific attributes as levers.
Click signals reward pages people choose, and you can't fake choosing
Yes, the dominant search engine uses clicks in ranking. That was described in court testimony before this week, and the documentation adds field names to it. In practice, click re-ranking means three things you can work on honestly. The title and description have to win the click against the results around them. The page has to end the search, so the visitor doesn't go back and pick another result. And the visit has to be long enough to look like the person found what they wanted, which is roughly what an attribute called lastLongestClicks suggests is tracked.
For a local accountant, that looks like a page for 'self-assessment help' that shows the fee range, the deadline and a booking link above the fold, instead of 800 words about the history of tax returns. The person gets the answer, books or calls, and has no reason to go back to the results. That's the click being rewarded for the right reason.
Is it worth trying to boost click-through artificially? No. Click farms, paid traffic bots and micro-task schemes that pay people to search and click are being pitched hard this week, and they're a bad bet for a client's domain. Systems that record clicks also record patterns, and a sudden rush of clicks from accounts with no history looks like nothing a real customer does. A demotion lands on your client's business, not on the vendor who sold the scheme.
The honest lever is volume of real intent. Get more people who already want this business to search for it and click. Someone typing the florist's name already knows where they're going, clicks the right result and stays. You can't manufacture that click, but you can create more of the people who make it.
Site-level authority means one strong page can't carry a weak site
A siteAuthority attribute changes the unit of work. If some measure of quality is held for the whole domain, then the 40 thin location pages, the blog section nobody has touched since 2021 and the near-duplicate service pages aren't harmless clutter. They're part of what the site gets judged on. Your one excellent page for 'emergency dentist' sits on a site that also hosts 60 pages no visitor would choose.
So what is site authority, and do domain-level scores matter? The documents show a site-level authority value exists. They don't show how it's calculated or weighted, and it isn't the same thing as the third-party domain scores sold by SEO tools. Those are outside estimates. The practical takeaway holds either way. Clean up the whole site, not just the money pages.
Deciding which pages deserve to exist is the hard part, and keyword intent mapping is the method we'd use. Map each page to one real search intent. Where two pages serve the same intent, merge them. Where a page serves no intent at all, remove it or noindex it. Doorway pages built to catch 'plumber in' plus every nearby village are the first to go, because they exist for the crawler and not for a person.
Brand demand is the signal small business clients can actually build
A dental practice in Bristol won't beat national health sites for 'tooth pain causes'. It doesn't need to. What it can grow is the number of people searching for it by name, visiting its site directly and coming back. Those are the behaviours that click data, site-level reputation and entity recognition all line up behind, and they're the ones a small budget can move.
Branded queries look like this, and they're worth tracking one by one.
- Business name plus service, such as 'Harbourside Dental Invisalign'
- Business name plus town, such as 'Stem and Stone florist Ludlow'
- Business name plus 'reviews', such as 'Okafor and Lane accountants reviews'
Each one is a person who already knows the business and wants something specific from it. They click the right result, they land, and they rarely go back to look for anyone else. That's the cleanest click signal there is, and it comes from somebody who had already decided before they searched.
How does a small local business build this when it can't outspend big sites? By being remembered by the people it already reaches. Past customers, enquiries that didn't book, people who signed up at the counter. Each reminder that puts the name in front of them, with a reason to visit, makes a branded search more likely next time they need the service. That's not an algorithm trick. It's how a local business has always grown, now with a trail you can see in search data.
It also fits what's landing in results this month. With AI-generated summaries now appearing above the links for many generic questions, as we covered in the May 2024 AI search summaries rollout, clicks on informational queries are under pressure. Branded queries hold up better, because a person searching for a business by name wants that business, not a summary. The less your client depends on winning generic informational terms, the less a summary box takes from them.
So should you change the SEO work you're selling, or is this just noise? The attribute names are noise until someone measures them. The direction isn't. Every signal people are discussing rewards a business with real demand behind it, and the rest of this article covers the most direct way an agency can build that demand for a client, starting with the list they already own.
Email is the most direct way to turn a list into branded searches and return visits
A florist with 1,800 past customers on her list has something no title tag gives her. She can put her shop's name in front of 1,800 people who already said yes, on the day she picks. Every send does three jobs. It reminds people the business exists. It gives them a reason to visit the site now. And it builds the habit of looking the business up by name when the need comes back. Someone who got a Mother's Day email in March types 'Stem and Stone Ludlow' in May instead of 'florist near me'. That's a branded search, a click on the right result and a visit that ends the search.
So does email marketing affect SEO? Not directly. No attribute in the documents says 'this site sends a newsletter'. What email moves is the behaviour the documents point at, which is people choosing the site, arriving direct and coming back. It's also owned reach. A change to the results page, a summary box above the links or a demotion elsewhere on the site doesn't take the list away.
Three things decide whether it works, and each one fails quietly if you skip it.
- A clean list. Dead addresses, old role accounts and people who signed up five years ago bounce or complain, and that damages the sending domain before the campaign has a chance. Verify addresses before the first send.
- Mail that reaches the inbox. A brand-new sending domain pushed straight into a full-list send lands in spam. It needs warmup, and volume needs pacing by provider.
- Links to pages worth landing on. Sending 1,800 people to a thin page wastes the visit. Link to the booking page, the seasonal offer or a genuinely useful article, so the visit looks like a person who found what they came for.
We built SendTo to handle the first two for agencies. You run it under your own brand, and it verifies addresses before sending, warms up new sending domains automatically, spreads campaign volume across several brand-matched sending domains and paces dispatch per provider and domain. Its reports exclude security scanner opens, so the numbers you show clients come from people. What it won't do is make a dull email worth opening or a weak landing page worth staying on. That part is still your work. Plans are on the pricing page.
Measure the link between sends and search demand
One send proves nothing. A pattern over two or three months is persuasive, and the method takes about an hour a month per client.
- Log every send date and the page each email linked to.
- In the client's search console, filter queries containing the business name and export daily impressions and clicks.
- In analytics, export daily direct sessions and returning visitors.
- Compare the three days after each send with the same weekdays in weeks with no send.
- Judge the email side on real human opens and clicks, not raw totals.
Expect noise. A mention in the local paper, a bank holiday or the start of wedding season will move branded searches on their own, so note anything unusual next to the dates. The last step matters more than it looks. Security scanners at company mail gateways open messages and follow links automatically, which can make a send look twice as engaged as it was. If you're correlating search demand with inflated clicks, the whole comparison is off. Our guide to tracking real human open rates covers how to separate the two.
Content earns its keep when it gets distributed, not just published
A 1,200-word post written to rank for 'how much does Invisalign cost' and then left alone is exactly the work this leak devalues. If clicks, site-wide quality and entity recognition all sit in the record, a page nobody reads isn't neutral. It's one more weak page on the domain. Content that gets sent, shared and picked up does the opposite. Readers arrive and stay, some search for the practice afterwards, and a local site quotes it and names the practice.
The loop is simple. Publish one piece that answers a question the client's customers really ask. Send it to the list within a week, with one line on why it matters to them. Offer a version to a trade or local publication with credit and a link back, which is the heart of content syndication. Then watch, over the following month, for branded mentions and for queries that pair the client's name with the topic.
The trade-off is time. A piece you distribute takes more hours than one you just publish, so you'll produce fewer. One a month done this way beats four left to sit. A client used to 'four blogs a month' will notice the count drop, so tell them why before the invoice does.

Rewrite the SEO retainer around demand, not rank positions
Clients still want to know where they rank, so keep rankings in the report. Just stop making them the product. A retainer built on demand has four parts, and each one maps to something the leaked documents point at.
- Site-quality cleanup. A quarterly crawl, thin pages merged or removed, indexing and speed fixes.
- One content piece a month, written to be sent and syndicated, not only to rank.
- A fixed email cadence, such as two sends a month to the client's own list, each linking to a page on their site.
- Demand reporting. Branded search impressions and clicks, returning visitors and email-driven sessions, with rankings for a short list of commercial terms alongside.
What should you report instead of keyword positions? Lead with the three demand numbers, show the trend against your baseline, then show positions underneath. A dental practice owner understands 'more people searched for us by name this quarter' far faster than a table of 60 keywords. For pricing the package and keeping it renewing, our retainer strategy guide covers the structure in detail.
Should you tell existing clients about the leak? Yes, briefly, and before they read a scary headline. Here's a version you can adapt this week.
You may see headlines this week about leaked search engine documents. They list what the search engine records, but not how much any of it counts, so nobody outside the company can say exactly what changes. What they point to fits what we already do for you, which is getting people to choose your site, trust it and come back. From next month we'll add branded searches and returning visitors to your report, next to rankings, and we'll start a regular email to your customer list to grow both. Nothing on your site needs an urgent change.
What to do this week
The search ranking documents leak SEO story will keep running for weeks, and most of what gets written will be guesswork about attribute names. You don't need to wait for it to settle. These five steps hold up whatever the analysis turns up.
- Don't send panic reports. No attribute scorecards, no emergency audits. Send the short script above instead.
- Audit thin pages. Export every URL with 12 months of sessions, and list anything with almost no visits and no links to merge, improve or remove.
- Set a branded search baseline. Pull 16 months of impressions and clicks for queries containing each client's name.
- Start or restart a regular send to each client's list. Clean the list first, warm up any new sending domain, and link each email to a page worth landing on.
- Reframe the reporting. Put branded searches, returning visitors and email-driven sessions at the top, and rankings underneath.
None of this needs the leak to be fully understood. It needs you to sell what the documents keep pointing at, which is a business people know, search for and come back to.
If you recommend tools to agencies, SendTo's affiliate programme pays you 20% of the partner subscription each agency you refer pays us, every month they keep paying. It never includes a share of what the agency's own clients pay the agency. Attribution lasts 90 days from the first click, first touch wins, and commission is paid monthly by PayPal.
Join the SendTo affiliate programme

