TextGPT Review: clear idea, rough execution (66/100)

TextGPT is an SMS-based AI assistant with prepaid usage credits.

Field	Value
Domain	textgpt.net
Category	SMS-based AI assistant for text questions and image generation
Pricing	Usage-based; 12–24 USD
Crawl	11 pages crawled on 2026-10-08
Audience	People seeking AI help through ordinary SMS, including hikers and low-connectivity users

TextGPT communicates a practical proposition: use an AI assistant through ordinary SMS, with a clear starting path and prepaid credit pricing. Audience and messaging score 78/100 because the workflow, credit model, SMS segments, image tiers, and model choices are concrete. The strongest question for a buyer is fit: country, phone, carrier, and connectivity determine whether the service is usable.

Execution is less mature. Usability scores 72/100, while design execution scores 48/100 and performance scores 48/100. Severe mobile instability and blocking work are the stated performance costs. The score is also held back by limited proof path, undersized or overflowing mobile controls, skipped heading levels, and a confirmed soft-404 pattern in which arbitrary paths can resolve as the homepage with HTTP 200. Accessibility scores 68/100: the signup form labels all four inputs, but structural issues remain.

TextGPT makes its central promise clear: access AI through ordinary text messages without another app. The homepage claim, "Tap into the power of ChatGPT and AI image generation through text messages," is supported by SMS, hiking, and ChatGPT guides, plus prepaid $12 and $24 credit packs. The strongest positioning move is to own "AI that works wherever SMS works" while keeping the US/Canada limitation and independence from OpenAI beside signup. A named founding team and Product Hunt proof help, but the audience remains broad.

The homepage explains the offer and sends visitors to signup, but the trust task is less direct. This is a simulated cognitive walkthrough against crawl data, not live user testing, so the points below describe observed friction rather than confirmed user behavior. Company is a weak trust cue, while the supplied inventory exposes no customer or case-study destination. Pricing is present in homepage copy—$12 for 240,000 credits and $24 for 480,000—but no pricing destination appears in the supplied inventory. Mobile adds measurable friction: primary links are 40px high, the form is 400px wide inside a 390px viewport, and mobile interaction is delayed by 11.3s time to interactive. Add an explicit trust path, connect pricing to conversion, and fix the mobile geometry.

The homepage explains the offer and sends visitors to signup, but the trust task is less direct. Company is a weak trust cue, while the supplied inventory exposes no customer or case-study destination. Pricing is present in homepage copy—$12 for 240,000 credits and $24 for 480,000—but no pricing destination appears in the inventory. Mobile adds measurable friction: primary links are 40px high, the form is 400px wide inside a 390px viewport, and mobile interaction is delayed by 11.3s time to interactive. Add an explicit trust path, connect pricing to conversion, and fix the mobile geometry.

The signup form is a sound baseline: all four inputs have explicit labels, and sampled images have no missing-alt findings. The structural gaps are more consequential. The homepage has no main landmark, its outline jumps from h2 to h4 and h5, and the sampled pages report no skip link. CSS contains animations without a prefers-reduced-motion query, while PSI reports CLS values of 0.913 and 1.252 alongside unsized images. Add a main landmark and skip link, restore heading order, declare reduced-motion behavior, and reserve image space. Keyboard traversal and screen-reader announcements were not verified.

TextGPT has a recognizable landing-page structure, a 16px body baseline, and a valid width=device-width viewport. Supporting templates are less robust. At 390px, /login and /signup report 400px scroll width; login inputs are 39px high, the Login button is 32px, and input text is 14px. The homepage outline skips from h2 to h4 and h5. PSI records CLS of 1.252 on desktop and 0.913 on mobile, and flags unsized images. Fix form geometry and control dimensions first, then reserve image space and consolidate the reported 90 colors, 14 font sizes, and 9 radii into a smaller token scale.

TextGPT misses the mobile lab thresholds mainly through instability and main-thread work. Mobile LCP is 3.3s, CLS is 0.913, TBT is 1,682ms, and time to interactive is 11.3s; no CrUX field metrics were returned. The homepage transfers 1,820 KiB across 32 requests, including 859 KiB from Stripe and 174 KiB from Google Tag Manager, while PageSpeed estimates 338 KiB of unused JavaScript. Reserve space for the unsized /2phones.webp image, defer payment and analytics code until needed, and reduce initial JavaScript before retesting under the same mobile conditions.

TextGPT is strongest when it explains credits, SMS segments, model costs, and image tiers. The hiking guide's warning, "This is not a rescue service," is a useful example of clear product limits. The editorial debt is structure: the homepage contains 647 words in one extracted paragraph, several pages lack a descriptive H1, and the About page contains "unsupported new texting app" plus the typo "sofware engineer." Restore paragraph boundaries, add descriptive H1s, replace the unsupported claim with concrete language, and use the hiking guide's qualified wording consistently.

The sampled baseline is stable: homepage and sitemap URLs return 200, robots.txt permits crawling, HTTPS and host redirects are present, and server-rendered text matches the raw HTML. The confirmed risk is URL hygiene. The tested guaranteed-nonexistent path returns the homepage title with HTTP 200, so it can blur the indexable URL set. Fix the response status first, then connect Search Console to measure whether unintended URLs were indexed or received impressions. The supplied evidence does not quantify traffic impact.

The supplied evidence supports a competent-with-gaps editorial assessment. Prioritize the 647-word homepage block and heading structure, then correct the About-page typo and replace its unsupported superlative with concrete product language. The copy has useful specifics—credit economics, SMS segment billing, model choices, and image costs—but those details are harder to scan when sections collapse into long extracted blocks. This dimension is grounded in the supplied crawl metrics and findings; it does not claim a broader corpus quality beyond the 11-page sample.

Technical health is broadly sound in the 11-page sample: robots.txt allows crawling and declares the sitemap, seven sitemap URLs return 200, raw and rendered word counts align, HTTPS and www redirect to the non-www host, and HSTS is present. The priority fix is the soft-404 probe, which returns the homepage with status 200 instead of a real 404. /login and /login/ also both return 200 without a redirect, creating a smaller canonical-choice problem. Return a genuine 404 for unknown paths, choose one login URL, and add missing H1s to sampled public pages.

TextGPT is a credible fit for people who want AI assistance through ordinary SMS, including hikers and people in low-connectivity situations. Its clearest strengths are the concrete workflow, direct signup path, prepaid usage model, and product copy that explains credits, SMS segments, image tiers, and model choices.

Three fixes should lead the next pass. First, stabilize mobile layout and speed. Second, strengthen proof and trust cues around the signup decision; the supplied inventory did not show a customer or case-study destination. Third, correct URL handling for guaranteed-nonexistent paths. Prioritize mobile stability, trust evidence, and correction of the soft-404 response before expanding the proposition.

This review covers the applicable dimensions supported by the supplied crawl evidence, page rendering and performance observations, accessibility facts, content checks, and technical SEO signals. The crawl covered 11 pages on 2026-10-08. Scores reflect the public dimension table; this article does not add owner-only findings or re-score the site.

Dimensions 12 (Decision-support surfaces) and 13 (Review-content integrity) were excluded as not applicable. Google Search Console access was not connected, so no GSC evidence is included. See How SiteList scores for the scoring method and dimension definitions.