Last updated August 2026. This protocol reflects Google’s confirmed August spam update and the reported originality-ranking shift. Expect a refresh at the next confirmed core update.
Update: August 2026
Google rolled out its third spam update of 2026 between Aug. 18 and Aug. 21, according to the Search Status Dashboard, first reported by Search Engine Roundtable. Google described it only as a normal spam update and did not name link spam or site-reputation abuse as the target.
Separately, publishers reported a distinct, unconfirmed shift around Aug. 8, 2026. SEO Vendor and other trackers describe pages with original data, first-party testing, or named expertise pulling ahead of pages that mainly summarize other sources. Google has not confirmed this as a distinct core update, as of Sept. 2, 2026.
Treat the two events separately. The spam update is confirmed and complete. The originality shift is a pattern multiple trackers report, not a named Google update. Either way, the fix below works: give the page something it could not have without you.
What counts as first-hand evidence
First-hand evidence is proof a page could not have without direct contact with the subject: a number you calculated, a test you ran, a photo you took, or a person you can name. It is the opposite of a roundup that restates five other articles in different words.
This same test already runs inside generative engines. Aggarwal et al.’s GEO study (Princeton, IIT Delhi, Georgia Tech, and the Allen Institute for AI, ACM KDD 2024) found that adding statistics to a page lifted its visibility in generative answers by roughly 30% to 41%. Adding direct quotations lifted it by roughly 22% to 40%. The two systems reward the same thing for the same reason: proof beats paraphrase.
That overlap is the real reason to act now. You are not chasing two separate signals. You are removing the parts of a page any competitor could publish and replacing them with parts only you could.
Why the roundup format stops working
For years, the fastest way to build a money page was to read the top ten results and rewrite them in your own words. That format still ranks in plenty of niches. But it is exactly the shape of page both signals in this update seem to target: no original claim, no named source, nothing a reader could not get from three other tabs.
Retire the “here is what everyone else says” page format. Replace it with a page that states, plainly, what you tested, what you counted, or who you talked to. That is a content strategy decision, not a technical fix, and it is the one decision this update and generative citation behavior now agree on.
The five evidence types, and the fastest way to add each
A page does not need all five types below. It needs at least one per major claim, placed high enough that a reader, or a crawler, hits it before the summary.
| Evidence type | What it looks like on the page | Fastest way to add it |
|---|---|---|
| Original data | A number nobody else publishes: usage stats, a survey result, or a calculation from your own records | Pull one metric from a report your team already runs and put it in the first three paragraphs |
| First-party testing | A described process you actually ran, with the setup, the result, and what surprised you | Write up the last time your team used the product or method, even informally |
| Screenshots and visuals | An image that could only exist if you did the thing: a settings screen, a before-and-after photo, a signed result | Take one screenshot the next time you do the task and caption it with the date |
| Named expertise | A credentialed person’s name attached to one specific claim, not a generic byline | Ask a specific person on your team for one direct quote, attributed by name and title |
| Primary quotes | A direct statement from a customer, expert, or source you talked to yourself | Pull one line from a support ticket, sales call, or interview, with permission |
The 9-step retrofit protocol
Run this against every money page tied to a query where you compete on trust, not just keywords: comparison pages, buying guides, pricing pages, and “best X” roundups.
- Inventory your money pages. List every page built to convert: comparisons, buying guides, pricing pages, and “best X” roundups. Sort by traffic or revenue, not by publish date.
- Run the “only you” test on each one. Read the page and ask whether a competitor could publish this exact paragraph without changing a fact. If yes, it fails.
- Score each page against the five evidence types. Count how many of the five, original data, first-party testing, screenshots, named expertise, and primary quotes, actually appear. Most money pages score zero or one.
- Flag every paragraph that only restates other sources. Roundup language (“many users report,” “reviewers note”) is the tell. Mark each instance directly in the draft.
- Replace one flagged paragraph per section with one evidence type. Do not try to fix the whole page at once. One real number or one named quote per section beats five vague claims.
- Rewrite the page in an editor that scores the fix as you go. Surfer fits this step: its content editor scores structure and factual density in real time, so you can see the retrofit land before you publish.
- Attach a name and a credential to the new evidence, not just a byline at the bottom. A quote from “our team” scores lower than a quote from a named person with a stated role.
- Republish and request reindexing. Update the page’s modified date, resubmit it through Search Console, and confirm the new evidence sits above the fold, not buried under the old summary paragraphs.
- Track the shift in organic rank and AI citation rate together. Profound and Temso (from $89/mo, unlimited projects and users on every plan) both track citation share across engines. Either one tells you whether the retrofit moved the number, not just the rank.
Before and after: one page, retrofitted
Picture a typical “best invoicing software” comparison page targeting a generic small-business tool.
Before the retrofit, the page reads like this: “Most small businesses find the tool’s invoicing features easy to use. Reviewers say it integrates well with popular accounting software. Users on forums often mention responsive customer support.” Every sentence restates what other review sites already say. A competitor could publish the same paragraph.
After the retrofit, the same section opens with evidence a competitor cannot copy: a specific test the writer ran (a set number of invoices sent over a set number of weeks, timed from draft to paid), one measured result compared against a stated baseline, and one support interaction logged with a date and a response time. The claims stay modest and specific. Nothing is invented, and nothing claims a scale of testing the writer did not actually do.
The difference is not tone. It is provenance. The “after” paragraph could only exist if someone actually used the product and wrote down what happened.
Common mistakes when you retrofit evidence
- Adding a stat with no source. A number without a named origin reads as invented, even when it is not. State where the figure came from in the same sentence.
- Burying the evidence below the summary. If the roundup paragraph still opens the page, the retrofit did not work. Move the evidence up, not just in.
- Using “our team” instead of a name. Generic attribution scores lower than a named person with a stated role, on both the ranking side and the citation side.
- Retrofitting every page at once. A rushed pass that adds thin, unverifiable claims to fifty pages does more damage than a careful pass on five. Ship depth over volume.
What to do this week
Pick your three highest-traffic money pages. Run the “only you” test on each one, and retrofit one section with real evidence before you touch anything else. Give the change four to eight weeks, then check both your organic rank and your citation rate, not just one of the two.
This retrofit protocol is the execution step inside the broader three-step GEO playbook for getting cited in AI answers: track where you are cited, diagnose the gap, then execute with the evidence types above. For the full definition of citation share and how it is measured, see the GEO glossary. For how a GEO ranking like the one on this site scores platforms, see the methodology. The full comparison of tools that track citation share lives at /rankings/geo-tools.