Quick answer: when a home-service company moves to Hydra, we rewrite almost all of the content it arrives with. Not because the old copy was lazy — usually somebody worked hard on it — but because it was written for a search engine that has since changed its behavior, it cannot fill the architecture we build, and a meaningful share of it contains claims nobody can verify. Below is the reasoning, section by section, and the short honest list of the 1% we leave alone.
One thing first, because it matters more than the rest of the article. “99%” is the name of a policy, not a measurement. We did not run a study and find that 99.0% of inherited copy fails. It is the number I use internally to set the default: assume the content gets rewritten, and make the case for keeping a page rather than the case for replacing it. If you want a real measured figure in this post, there is exactly one kind, and it is always attached to a source.
The content you are migrating was written for a search engine that no longer exists
Most incoming copy is a fossil of keyword-density-era SEO. You can spot the shape of it without reading closely: a services page with a paragraph per service, a service-area page that lists towns, a set of city pages where one paragraph has been duplicated and the town name swapped, and headings written to hold a phrase rather than to answer a question.
That was a rational thing to build in 2016. It is not what earns anything now, and the cost of carrying it forward is not theoretical. Champion Air, one of the largest family-owned HVAC teams in the Phoenix Valley, came to us after a December 2025 “optimization” dropped their organic traffic roughly 75 percent — from about 4,500 visits a month to 1,200. The diagnosis in their own launch story is thin content plus a WordPress architecture that merged service and city pages and chased low-demand keywords, cannibalizing its own search performance.
That is what old content actually does to a business. It does not sit there neutrally waiting to be improved. It competes with the good pages you are about to build.
Answer engines extract. They do not skim.
A person skims. They tolerate a welcome paragraph, scroll past the stock photo, and find the sentence they came for. An assistant does not do that. It reads the page looking for something it can lift and attribute — a direct answer, an extractable definition, a specific fact.
On Aug 3–4, 2026 we ran the same problem past ChatGPT, Gemini and Perplexity to watch this happen. Asked nationally — why is my air conditioner freezing up — the assistants pulled from 56 sources (Gemini) and 17 (Perplexity), every one of them a national publisher, manufacturer, or large directory. Asked with a city attached, the citation pool collapsed to 18, 20 and 6, and local contractors started appearing by name.
The single most useful thing we found was how Perplexity justified one of those local picks:
“B&B Air Conditioning & Heating Service — offers 24/7 emergency AC repair in Annapolis and says frozen coils are a common cause of AC problems.”
Read the second clause again. The assistant cited them because they had said something quotable about frozen coils. Not because of a review count, not because of a directory listing. Because there was a sentence on their site an assistant could attribute.
Legacy service copy almost never has that sentence. It has “Our friendly, professional technicians are here for all your cooling needs.” An assistant reads that, finds nothing it can stand behind, and moves to the next source. The page was not penalized. It was read and discarded, which is worse, because nothing in your analytics will ever tell you it happened.
The old copy physically cannot fill the new architecture
This is the least philosophical reason and probably the most decisive one.
A Hydra build is silo-per-service and page-per-city. Each service line gets its own content architecture rather than a shared services page; each market gets a real destination rather than a name in a list. Fish Premier HVAC went live on Aug 3, 2026 with 427 pages in the launch sitemap — seven service silos, six Dallas–Fort Worth markets, and a 313-post editorial library published on day one rather than promised for later.
Now count what typically arrives from the old site: one services page, one service-area page, an about page, a contact page, and a handful of blog posts. Joe Rushing Plumbing, a Lubbock institution founded in 1948, came to us on a 38-page site. Their rebuild runs 500-plus pages.
There is no honest way to spread 38 pages of copy across 500. “Migrating the content” into that structure means shipping empty rooms — a beautiful silo with one thin paragraph in it, which is exactly the failure mode the architecture was built to fix. The architecture is a promise of depth. Depth has to be written.
Near-duplicate pages are a liability, and we will not carry them forward
Name-swapped city pages are the most common thing we are asked to preserve and the thing we are most consistent about refusing.
Twelve near-identical pages do not give you twelve chances to rank. They give search engines twelve versions of one document and no reason to prefer any of them, and they give an assistant twelve pages with nothing specific in any of them. First Choice Garage Doors arrived with the same problem one level up: the business was spread across more than one website with overlapping content, competing against itself in search, with several Google Business Profiles suspended or duplicated — all of it quietly capping local visibility right as they pushed into new markets across five branches.
If a city page is going to exist, it has to earn its existence with something true only about that city: the crews who work it, the code quirks, the neighborhoods, the drive time, the jobs you actually do there. That is not a rewrite of the old page. That is a new page, and it has to come out of a conversation with the operator.
Rewriting is when unverifiable claims get caught
This one is not a marketing argument. It is a risk argument, and it is the reason I will not let a migration skip the rewrite even when the copy reads fine.
When you take every sentence apart, you find things: licence numbers nobody can produce, an award badge from 2019, a “100% satisfaction guarantee” the service manager has never heard of, financing terms that expired, a rebate that ended two seasons ago, and a service-area list including towns the company stopped driving to years ago.
Our intake brief for trades and home-service companies is explicit about this, and the rules are worth quoting because they are the whole discipline in four lines:
“No invented prices, rebates, licenses, or guarantees operations cannot honor.” Every Google-verified business address and Business Profile link must be supplied by the operator — “never invent these.” “Do not list cities they will not travel to.” And when intake data conflicts with itself, the conflict gets surfaced under Gaps & conflicts — “do not invent a winner.”
Applied to a rewrite, that means every price, licence, certification, rebate, guarantee and service area on the new site traces back to something the operator gave us. If we cannot source it, it does not ship. That rule occasionally costs a client a nice-sounding line. It has never once cost them anything real.
E-E-A-T needs a named human, and legacy content is anonymous
Look at the byline on the content you are migrating. There usually is not one, or it says “admin.”
Experience, expertise, authoritativeness and trust are not a schema field you switch on. They are a claim that a specific, identifiable person stands behind what the page says. On our own builds every post resolves to a real person with a real bio and a real page — the author roster is derived from the team roster, so a post cannot quietly publish as nobody. Where the subject is regulated or safety-adjacent, it gets a reviewer as well as a writer, and a last-updated date that reflects an actual review rather than a cron job touching a timestamp.
None of that can be retrofitted onto inherited copy, because you cannot attach a named expert to paragraphs no expert has read. Attribution is a rewrite trigger by itself.
The 1%: what we keep, and keep exactly
If the policy had no exceptions it would be a sales pitch. It has four, and they are specific.
- The owner’s own story, in the owner’s own words. When a founder has written something genuine about how the company started or what it stands for, we publish it verbatim. Nobody on my team writes a better version of a first-person story than the person who lived it, and every attempt reads like it.
- Real testimonials, where permission exists. A customer’s sentence is theirs. We do not tighten it, we do not fix the grammar, and we do not publish it without permission — testimonial permission is its own line item in our intake brief for exactly this reason.
- Legal and compliance language operations must honor. Warranty terms, guarantee language, licence display requirements, privacy and terms, SMS consent. If operations has to live by the words, marketing does not get to improve them.
- Any page that is already working. If a page ranks, gets cited, or converts, we keep its URL and its substance. Preserving a URL is an atomic, SEO-preserving operation on our platform rather than a redirect somebody remembers to write, which means the honest answer to “will I lose what I have” is no — and the burden is on us to show which pages qualify before we touch anything.
Everything outside those four categories is in scope. That is what 99% means in practice.
Rewriting is not “more words”
The most common misread of this policy is that we are going to inflate everything. The opposite is usually true — plenty of rewritten pages come out shorter. What changes is structure and specificity.
- Answer first. The first paragraph answers the question in the heading. No welcome, no throat-clearing, no history of the company. If someone lands mid-decision from an assistant, they get the answer before they get the pitch.
- One job per page. A page that explains, compares, and sells at once does none of the three well. Doggone Good launched in Baton Rouge with a 13-article library sorted into Learn, Compare and Act, so each question has a page whose only job is that question.
- Specifics over adjectives. “Fast, reliable service” is a claim any competitor can copy in ten seconds. A named response window, a named brand, a named neighborhood, a real number the operator supplied — those are facts, and facts are what an assistant can quote.
- The operator’s vocabulary, not ours. If the crew calls it a changeout, the page says changeout. Copy that does not sound like the company is a liability the first time a homeowner calls.
What this does not promise
Two honest caveats, because a post arguing for rigor should be rigorous about itself.
Rewriting is not a fast-acting result. New content has to be published, crawled, and established before anything cites it. In our Aug 2026 probes, Still’s HVAC had gone live on Hydra the day before and did not appear in the Annapolis answers — which is exactly what should happen. Anyone promising fast AI citations from new copy is guessing.
Rewriting does not fix a business problem. If the phone gets answered badly, if the trucks are late, if the pricing is not real, better content will get more people to a worse experience faster. Content is the part of the machine we can build. It is not the whole machine.
What this asks of you
The uncomfortable part of this policy is that it puts work on the operator, and I would rather say that plainly than surprise you with it.
We cannot rewrite your site out of your old site. We can only rewrite it out of what is true about your company — the services you actually sell, the towns you actually drive to, the licences you can actually produce, the guarantees your service manager will actually honor, the people whose names go on the pages. That is a real intake, and the quality of the site tracks it almost exactly.
What you get back is a site where every page has a job, every claim has a source, every byline has a person, and every answer is one an assistant can quote. That is worth more than a faithful copy of content built for a search engine that stopped behaving that way years ago.
Related reading
- How consumers search with AI, the Aug 2026 citation probes quoted above.
- Content intelligence, how we decide what to write before we write it.
- How generative search reads a site.
- Hydra launches, including Champion Air, Joe Rushing, Fish Premier, and Doggone Good.
- Get Started if you want this done on your own site.