The 9pm Test, Houston: what 142 personal injury firm websites show a Spanish speaker after hours
On 15 September 2026 we read the public websites of 148 personal injury law firms in the Houston metro; 142 were readable. 23 have a page actually written in Spanish. 43 have a Spanish page that reads as machine-translated, 21 have a translation widget and nothing else, and 55 have no Spanish page at all. 38 of 142 both say they take contact after hours and name a Spanish speaker. 10 of the 107 forms we could read ask a single qualifying question. No firm scored above 5 of 7. Nothing was called and no form was submitted.
Crawled 15 September 2026 · Published 15 September 2026 · Method version 1.0 · 148 firms on the list, 142 readable, 6 unreachable
This is the Houston run of the 9pm Test, a dated, reproducible check of what a person looking for help at 9pm actually finds on a business's public pages, scored across seven signals. The question it asks in personal injury is a narrow one: if a Spanish-speaking accident victim finds a Houston firm at 9pm, what do the firm's own pages promise them? In the Houston metro, 30.2% of people aged 5 and over speak Spanish at home (US Census Bureau, ACS 2024 1-year, table S1601), so the answer is not a niche.
Everything below is a count. No firm is named for failing, here or anywhere else we publish. Sites that could not be read are reported as unreachable, never as failures, and a signal that could not be judged is reported as unknown, never as a fail.
What we found
The seven signals, side by side. Each bar is the whole list of 148 firms; the grey part is what could not be judged from public pages and is never counted against a firm.
The 9pm Test, Houston, TX, crawled 15 September 2026. Pass, fail and unknown out of 148 firms. Our data
Signal 1: the Spanish page
The engine opens the home page, follows any Spanish link it finds (an hreflang tag, an /es/ or /espanol/ path, a Spanish subdomain, a link whose text says Español), and if nothing is linked it probes /es/ and /espanol/ once each. A home page that is itself in Spanish counts as the Spanish page. A page counts as written rather than translated only if it shows none of the tells below.
| What the site has | Sites | Share of readable |
|---|---|---|
| A page actually written in Spanish | 23 | 16% |
| A Spanish page that reads as machine-translated | 43 | 30% |
| A translation widget and nothing else | 21 | 15% |
| No Spanish page at all | 55 | 39% |
| Site unreachable, not counted either way | 6 |
142 readable sites. A site is counted once, in the best category it reached.
The tells that turned a Spanish page from written into translated. A page can show more than one, and one strong tell is enough to fail the signal.
| Tell seen on the Spanish page | Sites |
|---|---|
| the voice slides between usted and tú on the same page | 28 |
| the page tells the browser it is in English (lang attribute not es) | 17 |
| the navigation is still in English (Home, Contact, Blog) | 13 |
| the page is served through a machine-translation service | 9 |
| the Spanish page exists but nothing on the home page links to it | 6 |
| a literal calque from English (for example 'submit' rendered word for word) | 5 |
| a large share of the page is still English | 2 |
| the page title is still in English | 1 |
Counted on every Spanish page found, including pages that passed on other grounds. The usted and tú test ignores testimonial quotes.
Translation widgets and plugins found in the home page source. GTranslate and Google Translate rewrite the English page on the fly; WPML, Polylang and TranslatePress can hold a written translation or a machine one, so their presence alone is not a fail.
| Translation technology on the home page | Sites |
|---|---|
| GTranslate | 21 |
| Google Translate | 11 |
| WPML | 8 |
| Polylang | 8 |
| TranslatePress | 6 |
| Weglot | 3 |
| Bablic | 1 |
Signal 2: the phone number
On the Spanish page, or the home page when there is none, the engine looks for a tel: link, which is the only thing that makes a number dial when tapped on a phone. This is the one signal most sites pass: 132 of 142. Of the 10 that fail, 8 show the number as plain text and 2 show no number in the page text at all, which usually means it is drawn by a script or set inside an image.
Signals 3 and 7: the form, what it promises, and what it asks
The engine finds the best lead form without submitting it, preferring a form on the Spanish page, then a Spanish contact page, then any contact page, then the home page, and reads embedded third-party forms when they load from a public URL. It records the field labels, the language they are in, and whether any sentence near the form or on a linked thank-you page says what happens after you press send.
| The form | Sites | Share |
|---|---|---|
| Sites where a lead form could be read | 107 | 75% |
| Forms that sit on the Spanish page | 50 | 47% |
| Form fields labelled in Spanish | 39 | 36% |
| Form fields labelled in English | 67 | 63% |
| Form fields mixing both | 1 | |
| Any sentence near the form or on a thank-you page saying what happens next | 17 | 16% |
| That sentence in Spanish | 9 | |
| That sentence in English | 8 | |
| Signal 3 pass: Spanish form and a Spanish promise | 9 | 8% |
Shares in the first row are of 142 readable sites; the rest are of the 107 sites where a form could be read. 35 readable sites had no form in their page source, usually because it is drawn by a script, and are counted as unknown.
Where the 98 forms fail signal 3: 56 on sites with no Spanish page, where the form is in English by definition, 30 in Spanish with nothing near them or on a thank-you page saying what happens next, 10 on the Spanish page with fields in English, 1 on sites whose Spanish page has no form of its own, only the English page does, and 1 mixing English and Spanish fields on the Spanish page.
Signal 7 asks a different question of the same form: does it ask anything about the accident before the enquiry reaches the intake desk? 10 of 107 forms do. The rest ask for a name, a phone number and an email, and nothing else.
| Qualifying question asked | Forms |
|---|---|
| injured | 5 |
| at fault | 3 |
| date of accident | 3 |
| medical treatment | 2 |
| already have an attorney | 1 |
A form can ask more than one. Counted in English or Spanish.
Signal 4: after hours, and who answers
The engine reads the home, Spanish and contact pages for a line about hours (24 hours, around the clock, after hours, horas, disponible) and, separately, for a sentence that says a Spanish speaker answers (hablamos español, se habla español, atención en español). The word Spanish on its own, as a link or a label, does not count. The signal passes only when both halves are on the same site. This measures what the site claims, not whether anyone picks up.
| What the pages say | Sites | Share of readable |
|---|---|---|
| Says it takes contact around the clock or after hours | 66 | 46% |
| Lists business hours and says nothing about after hours | 9 | |
| States no hours and nothing about after hours | 67 | |
| Says somewhere that a Spanish speaker answers | 58 | 41% |
| Uses the word Spanish or español only as a link or label | 25 | |
| Claims after-hours contact but names no Spanish speaker | 28 | |
| Names a Spanish speaker but says nothing about after hours | 20 | |
| Has a chat widget in the page source | 52 | 37% |
| Signal 4 pass: both halves on the same site | 38 | 27% |
Chat widgets found in the page source. A widget is neither a pass nor a fail on its own; it is recorded because it is often the only thing that answers at 9pm.
| Chat widget | Sites |
|---|---|
| Intaker | 15 |
| Ngage | 12 |
| Juvo Leads | 8 |
| ApexChat | 4 |
| Client Chat Live | 4 |
| Birdeye | 3 |
| Smith.ai | 2 |
| tawk.to | 2 |
| Lawmatics | 1 |
| Crisp | 1 |
| LiveChat | 1 |
| Tidio | 1 |
Signal 6: paid-acquisition markers
The engine searches the page source for a Google Ads conversion or remarketing tag, a Meta Pixel, Microsoft UET, Local Services badge markup, and dynamic number scripts from CallRail, CallTrackingMetrics, WhatConverts or Invoca. This is a proxy for set up to buy traffic, not proof of live spend. When only a tag manager is present the signal is unknown, because a tag manager can load all of the above out of sight.
| Marker in the page source | Sites |
|---|---|
| Meta Pixel | 37 |
| CallRail | 36 |
| Google Ads conversion tag | 27 |
| Google Ads remarketing | 5 |
| WhatConverts | 4 |
| Microsoft UET | 4 |
| LSA badge | 1 |
| CallTrackingMetrics | 1 |
A site can carry several markers.
| Paid acquisition, summary | Sites | Share |
|---|---|---|
| Sites whose scripts could be read and judged | 112 | |
| Signal 6 pass: at least one marker present | 68 | 61% |
| No marker and no tag manager | 44 | 39% |
| Only a tag manager, which can load markers out of sight (unknown) | 30 | |
| Google Tag Manager present at all | 81 | 57% |
| Any call-tracking script | 41 | 29% |
Shares in the pass and fail rows are of the 112 sites that could be judged; the rest are of 142 readable sites.
Signal 5: the Facebook reply badge
Unknown for all 148 firms. Facebook does not put the "typically replies" badge in a Page's static HTML for most Pages, and the method records pass only when the badge is actually seen. Unknown is never counted as a fail, which is why the practical maximum on this run was 6 of 7 rather than 7.
How many signals each firm passed
Signals passed out of 7, across the 142 readable Houston sites. Green would mark 6 or 7; nobody is there. Our data
| Signals passed | Sites | Share of readable |
|---|---|---|
| 0 of 7 | 8 | 6% |
| 1 of 7 | 41 | 29% |
| 2 of 7 | 52 | 37% |
| 3 of 7 | 31 | 22% |
| 4 of 7 | 8 | 6% |
| 5 of 7 | 2 | 1% |
| 6 of 7 | 0 | 0% |
| 7 of 7 | 0 | 0% |
| Median | 2 of 7 |
8 readable sites passed nothing at all. 88 of 142 fail both the Spanish page and the after-hours signal together, which is the combination that describes a Spanish speaker at 9pm most exactly: no page written for them, and nobody named who would answer them.
Firms that passed
None. No firm in Houston scored 6 or 7 of 7 on 15 September 2026. The Facebook reply badge returned unknown for every firm, so the practical ceiling from public pages was 6, and nobody reached it. 2 firms scored 5 of 7, every site signal but one. The most common score among readable sites was 2 of 7 and the median was 2.
We name firms only when they pass. This section will carry the name and a link of the first Houston firms to reach 6, and we will send each of them the result first.
What the sites are built on
| Technology in the page source | Sites |
|---|---|
| WordPress | 120 |
| Google Analytics | 84 |
| Google Tag Manager | 81 |
| Gravity Forms | 54 |
| Elementor | 48 |
| Contact Form 7 | 40 |
| Cloudflare | 34 |
| WPForms | 16 |
| FindLaw | 15 |
| Martindale-Nolo | 11 |
| Divi | 11 |
| WPBakery | 7 |
| HubSpot | 6 |
| Justia | 5 |
| Ninja Forms | 5 |
| GoDaddy Website Builder | 4 |
| Webflow | 3 |
| Formidable | 3 |
Detected from the home, Spanish and contact pages of 142 readable sites. A site can carry several. Platform version strings are left out.
What this means for a Spanish-speaking accident victim at 9pm
Put the numbers in the order the person meets them.
She was hit this morning. She has been to urgent care, the kids are asleep, and at 9pm she picks up her phone and searches in Spanish. She taps a Houston firm. On 55 of the 142 sites we could read, nothing on the page is in her language. On another 21 the only Spanish is a translate button that rewrites the English page in front of her, with the mistakes that come with that. On 43 more there is a Spanish page, but it slides between usted and tú, or the menu still says Home and Contact, or it tells her browser it is in English. On 23 sites, one in 6, somebody wrote the page for her.
Say she stays. The number usually works: 132 of 142 sites will dial when she taps it. But it is 9pm. 66 of these sites say they take contact around the clock; only 38 of them also say, anywhere, that a Spanish speaker answers. On the other 28 she is dialling a promise made in English.
Say she would rather write than call. Of the 107 forms we could read, 39 are labelled in Spanish. 17 tell her what happens after she presses send, and 9 of those say it in Spanish. 10 ask her anything about the accident. So on most sites she types her name and number into an English box, gets no idea who will call her or when, and goes back to the search results.
That is the whole finding: the click works, the page does not, and the person is not named. None of it takes a phone call to see, and any firm can check its own pages tonight in about fifteen minutes with the method. Owners who raise the Spanish-speaking market in lawyers' forums tend to say the same thing: ads with nobody to answer in Spanish are money on fire. This is what the pages say before a dollar of ads is spent.
Sites we could not read
6 of 148 home pages could not be fetched with an ordinary browser request and are counted as unreachable, never as failures. They are left out of every share on this page.
| Why the home page could not be read | Sites |
|---|---|
| bot wall (captcha or JS challenge served instead of the page) | 5 |
| HTTP 403 | 1 |
Method, in short
- Source list. 148 personal injury firm websites in the Houston, TX metro, compiled from public directories and firm pages.
- How they were read. Plain GET requests with a browser user agent, a twelve second timeout, eight sites at a time and one request at a time per site, with a pause between requests to the same site. At most nine pages per firm: the home page, the Spanish page, a contact page, embedded form documents and a thank-you page when one is linked.
- What was not done. No call, at 9pm or any hour. No form submitted. No message sent. No accident invented. No login, no private group, no paid data. The Facebook Page was read only with a public request and only where the firm links one.
- Scoring. Each of the seven signals is pass, fail or unknown, with the evidence line, the URL and the date. Unknown is never a fail. The score is the number of passes out of seven.
- Names. Firms are named only when they score 6 or 7. Everything else is a count.
- Reproducibility. Every check on this page can be repeated by hand from the outside using the published method. The Houston and Los Angeles runs use the same code, the same date and the same rules, so the two pages can be read against each other. The Los Angeles run is at /research/9pm-test-los-angeles.
Download the data. data.csv holds every count on this page: the seven signals, the Spanish page categories and tells, the form, after-hours and paid-acquisition breakdowns, the score distribution and the technologies. Aggregate only; the only firm names in it are the ones that passed, and this run has none.
Cite as: The 9pm Test, Houston, TX, crawled 15 September 2026, Wealth Marketing AI, https://wealthmarketing.ai/research/9pm-test-houston. Method: https://wealthmarketing.ai/9pm-test.
Request your firm's result
Every firm on the list has a seven-line scorecard in this dataset, with the evidence and the date beside each line. We do not publish it. If your firm is in Houston and you want yours, pick a time here or reply to the email that brought you to this page, and we will send it before any call, whether or not we ever speak. If you would rather not wait, paste your site into the free tool at app.wealthmarketing.ai/tools/9pm-test, or run the seven checks by hand tonight.
Your firm's seven signals, dated, before any call.
Book a time and we will send your Houston scorecard first. Public pages only. Nothing called, nothing submitted.
Pick a time Or reply to the email that brought you here.Frequently asked questions
How many personal injury firms in Houston have a real Spanish page?
23 of 142 readable sites, or 16%, on 15 September 2026. A further 43 had a Spanish page that reads as machine-translated, 21 had a translation widget and nothing else, and 55 had no Spanish page at all.
What did the 9pm Test check on each site?
Seven public signals, each scored pass, fail or unknown with the line of evidence and the date: a page actually written in Spanish; a phone number that dials when tapped; a Spanish form that says what happens after you send it; a statement that someone takes contact after hours and that a Spanish speaker is there; the reply badge on the public Facebook Page; paid-acquisition markers in the page source; and whether the form asks one qualifying question about the accident.
Did you call the firms or submit their forms?
No. Every site was read with ordinary GET requests and a browser user agent, one page at a time, with a twelve second timeout. No number was dialled, no form was submitted, no message was sent, no accident was invented and nothing behind a login was touched.
Which Houston firms passed?
No firm in Houston scored 6 or 7 of 7 on 15 September 2026. The Facebook reply badge returned unknown for every firm, so the practical ceiling from public pages was 6, and nobody reached it. The highest score was 5, reached by 2 firms. We name firms only when they pass; the first Houston firms to reach 6 will be named on this page.
Why is the Facebook reply badge unknown for every firm?
Because Facebook does not put the 'typically replies' badge in a Page's static HTML for most Pages. The method records pass only when the badge is actually seen and unknown otherwise, and unknown is never counted as a fail. That is why the practical maximum on this run was 6 of 7.
How can my firm get its own result?
Book a time at wealthmarketing.ai/booking, or reply to the email that brought you here, and we will send your firm's seven signals with the evidence line and the date, whether or not we ever speak. You can also run the same seven checks yourself tonight from the method page, or paste your site into the free tool at app.wealthmarketing.ai/tools/9pm-test.
Sources
- Ours Wealth Marketing AI, The 9pm Test, Houston, TX, crawled 15 September 2026. 148 personal injury firm websites, 142 readable, public pages only. Method: https://wealthmarketing.ai/9pm-test. Data: data.csv.
- Primary US Census Bureau, American Community Survey 2024 1-year estimates, table S1601 (Language Spoken at Home). Spanish spoken at home, population 5 years and over, Houston, TX metro: 30.2%. data.census.gov, table S1601
The 9pm Test, the method · The Los Angeles run · Spanish-first case acquisition in Houston · Law firm intake statistics, 2026